CMakeLists.txt

cmake_minimum_required(VERSION 3.20)

project(AudioCpp LANGUAGES C CXX)

set(AUDIOCPP_VERSION "dev" CACHE STRING "audio.cpp version string")

set(CMAKE_C_STANDARD 11)
set(CMAKE_C_STANDARD_REQUIRED ON)
set(CMAKE_C_EXTENSIONS OFF)
set(CMAKE_CXX_STANDARD 17)
set(CMAKE_CXX_STANDARD_REQUIRED ON)
set(CMAKE_CXX_EXTENSIONS OFF)
set(CMAKE_RUNTIME_OUTPUT_DIRECTORY ${CMAKE_BINARY_DIR}/bin)
set(CMAKE_LIBRARY_OUTPUT_DIRECTORY ${CMAKE_BINARY_DIR}/bin)

option(AUDIOCPP_DEPLOYMENT_BUILD
    "Compile model specs into the runtime for self-contained deployments"
    OFF)

# Model specs remain editable runtime data in normal builds. The converter always
# carries the source catalog so the conversion binary is portable. Deployment builds
# additionally compile the catalog into engine_runtime as a fallback for CLI/server.
#
# Each spec becomes one C++ string, and MSVC caps a single string literal at 16380
# single-byte characters (error C2026) before adjacent literals are concatenated --
# qwen3_tts.json is 16.5 KB on its own. So split each spec into adjacent literals,
# cutting at newline boundaries so no chunk splits a UTF-8 sequence and every chunk
# stays plain JSON text.
function(audiocpp_model_spec_literal out json)
    set(AUDIOCPP_SPEC_LIMIT 16000)
    set(AUDIOCPP_SPEC_REST "${json}")
    set(AUDIOCPP_SPEC_LITERALS "")
    while(TRUE)
        string(LENGTH "${AUDIOCPP_SPEC_REST}" AUDIOCPP_SPEC_LEN)
        if(AUDIOCPP_SPEC_LEN LESS_EQUAL AUDIOCPP_SPEC_LIMIT)
            string(APPEND AUDIOCPP_SPEC_LITERALS "R\"AUDIOCPP_SPEC(${AUDIOCPP_SPEC_REST})AUDIOCPP_SPEC\"")
            break()
        endif()
        string(SUBSTRING "${AUDIOCPP_SPEC_REST}" 0 ${AUDIOCPP_SPEC_LIMIT} AUDIOCPP_SPEC_HEAD)
        string(FIND "${AUDIOCPP_SPEC_HEAD}" "\n" AUDIOCPP_SPEC_CUT REVERSE)
        if(AUDIOCPP_SPEC_CUT LESS 1)
            # A spec with no newline in the first 16 KB: cut at the limit.
            set(AUDIOCPP_SPEC_CUT ${AUDIOCPP_SPEC_LIMIT})
        endif()
        string(SUBSTRING "${AUDIOCPP_SPEC_REST}" 0 ${AUDIOCPP_SPEC_CUT} AUDIOCPP_SPEC_CHUNK)
        string(SUBSTRING "${AUDIOCPP_SPEC_REST}" ${AUDIOCPP_SPEC_CUT} -1 AUDIOCPP_SPEC_REST)
        string(APPEND AUDIOCPP_SPEC_LITERALS "R\"AUDIOCPP_SPEC(${AUDIOCPP_SPEC_CHUNK})AUDIOCPP_SPEC\" ")
    endwhile()
    set(${out} "${AUDIOCPP_SPEC_LITERALS}" PARENT_SCOPE)
endfunction()

file(GLOB AUDIOCPP_MODEL_SPEC_FILES CONFIGURE_DEPENDS "${CMAKE_CURRENT_SOURCE_DIR}/model_specs/*.json")
set(AUDIOCPP_BUILTIN_MODEL_SPEC_ENTRIES "")
set(AUDIOCPP_CONVERTER_MODEL_SPEC_ENTRIES "")
set(AUDIOCPP_NATIVE_PACKAGE_SPEC_ENTRIES "")
foreach(AUDIOCPP_MODEL_SPEC_FILE IN LISTS AUDIOCPP_MODEL_SPEC_FILES)
    get_filename_component(AUDIOCPP_MODEL_SPEC_FAMILY "${AUDIOCPP_MODEL_SPEC_FILE}" NAME_WE)
    file(READ "${AUDIOCPP_MODEL_SPEC_FILE}" AUDIOCPP_MODEL_SPEC_JSON)
    audiocpp_model_spec_literal(AUDIOCPP_MODEL_SPEC_LITERAL "${AUDIOCPP_MODEL_SPEC_JSON}")
    string(APPEND AUDIOCPP_CONVERTER_MODEL_SPEC_ENTRIES
        "    {\"${AUDIOCPP_MODEL_SPEC_FAMILY}\", ${AUDIOCPP_MODEL_SPEC_LITERAL}},\n")
    string(APPEND AUDIOCPP_NATIVE_PACKAGE_SPEC_ENTRIES
        "    {\"${AUDIOCPP_MODEL_SPEC_FAMILY}\", ${AUDIOCPP_MODEL_SPEC_LITERAL}},\n")
    if (AUDIOCPP_DEPLOYMENT_BUILD)
        string(APPEND AUDIOCPP_BUILTIN_MODEL_SPEC_ENTRIES
            "    {\"${AUDIOCPP_MODEL_SPEC_FAMILY}\", ${AUDIOCPP_MODEL_SPEC_LITERAL}},\n")
    endif()
endforeach()
file(MAKE_DIRECTORY "${CMAKE_CURRENT_BINARY_DIR}/generated")
file(GENERATE
    OUTPUT "${CMAKE_CURRENT_BINARY_DIR}/generated/model_specs.inc"
    CONTENT "${AUDIOCPP_BUILTIN_MODEL_SPEC_ENTRIES}")
file(GENERATE
    OUTPUT "${CMAKE_CURRENT_BINARY_DIR}/generated/gguf_converter_model_specs.inc"
    CONTENT "${AUDIOCPP_CONVERTER_MODEL_SPEC_ENTRIES}")
file(GENERATE
    OUTPUT "${CMAKE_CURRENT_BINARY_DIR}/generated/native_package_specs.inc"
    CONTENT "${AUDIOCPP_NATIVE_PACKAGE_SPEC_ENTRIES}")

if (NOT CMAKE_CONFIGURATION_TYPES AND NOT CMAKE_BUILD_TYPE)
    set(CMAKE_BUILD_TYPE RelWithDebInfo CACHE STRING "Build type" FORCE)
endif()

if (NOT MSVC)
    set(AUDIOCPP_DEBUG_OPT_FLAGS "-O3 -g")
    set(CMAKE_C_FLAGS_DEBUG "${AUDIOCPP_DEBUG_OPT_FLAGS}" CACHE STRING "Optimized debug C flags" FORCE)
    set(CMAKE_CXX_FLAGS_DEBUG "${AUDIOCPP_DEBUG_OPT_FLAGS}" CACHE STRING "Optimized debug C++ flags" FORCE)
endif()

set(ENGINE_DEFAULT_ENABLE_CUDA OFF)
set(ENGINE_DEFAULT_ENABLE_HIP OFF)
set(ENGINE_DEFAULT_ENABLE_VULKAN OFF)
set(ENGINE_DEFAULT_ENABLE_METAL OFF)
set(ENGINE_DEFAULT_ENABLE_LLAMAFILE ON)
set(ENGINE_DEFAULT_ENABLE_CUDA_GRAPHS ON)
set(ENGINE_DEFAULT_ENABLE_NATIVE_CPU ON)
if (APPLE)
    set(ENGINE_DEFAULT_ENABLE_METAL ON)
endif()
if (DEFINED GGML_CUDA)
    set(ENGINE_DEFAULT_ENABLE_CUDA ${GGML_CUDA})
endif()
if (DEFINED GGML_HIP)
    set(ENGINE_DEFAULT_ENABLE_HIP ${GGML_HIP})
endif()
if (DEFINED GGML_VULKAN)
    set(ENGINE_DEFAULT_ENABLE_VULKAN ${GGML_VULKAN})
endif()
if (DEFINED GGML_METAL)
    set(ENGINE_DEFAULT_ENABLE_METAL ${GGML_METAL})
endif()
if (DEFINED GGML_LLAMAFILE)
    set(ENGINE_DEFAULT_ENABLE_LLAMAFILE ${GGML_LLAMAFILE})
endif()
if (DEFINED GGML_CUDA_GRAPHS)
    set(ENGINE_DEFAULT_ENABLE_CUDA_GRAPHS ${GGML_CUDA_GRAPHS})
endif()

option(ENGINE_ENABLE_CUDA "Build ggml with CUDA backend support" ${ENGINE_DEFAULT_ENABLE_CUDA})
option(ENGINE_ENABLE_HIP "Build ggml with HIP/ROCm backend support" ${ENGINE_DEFAULT_ENABLE_HIP})
option(ENGINE_ENABLE_VULKAN "Build ggml with Vulkan backend support" ${ENGINE_DEFAULT_ENABLE_VULKAN})
option(ENGINE_ENABLE_METAL "Build ggml with Metal backend support" ${ENGINE_DEFAULT_ENABLE_METAL})
option(ENGINE_ENABLE_LLAMAFILE "Build ggml with llamafile SGEMM support" ${ENGINE_DEFAULT_ENABLE_LLAMAFILE})
option(ENGINE_ENABLE_CUDA_GRAPHS "Enable ggml CUDA graphs support" ${ENGINE_DEFAULT_ENABLE_CUDA_GRAPHS})
option(ENGINE_HIP_STRIX_HALO_OPTIMIZATIONS
    "Enable opt-in HIP optimizations validated only for Strix Halo (gfx1151)"
    OFF)
# Rockchip NPU (RK3588 & co): a model-local compute path, not a ggml backend —
# the Qwen3 TTS speech decoder runs a pre-compiled .rknn step graph through
# librknnrt. The engine links the board runtime; the models themselves are
# converted offline on x86 with rknn-toolkit2.
option(ENGINE_ENABLE_RKNN "Build the Rockchip NPU (librknnrt) decoder path" OFF)
option(ENGINE_ENABLE_NEON "Build opt-in NEON library and native model paths" OFF)
set(ENGINE_RKNN_INCLUDE_DIR "" CACHE PATH "Directory containing rknn_api.h (empty = use the system include path)")
set(ENGINE_RKNN_LIBRARY "rknnrt" CACHE STRING "librknnrt to link (path or library name)")
option(ENGINE_ENABLE_NATIVE_CPU "Build ggml CPU kernels with native host ISA flags" ${ENGINE_DEFAULT_ENABLE_NATIVE_CPU})
option(ENGINE_ENABLE_OPENMP "Build host code with OpenMP support" ON)
option(AUDIOCPP_STATIC_ESPEAK "Build and statically link GPL-3.0-or-later eSpeak-ng; keep data separate" OFF)
option(AUDIOCPP_BUILD_NATIVE_MODEL_MANAGER
    "Build native model-manager tools and server download/install support"
    OFF)
include("${CMAKE_CURRENT_SOURCE_DIR}/app/server/server_frontends.cmake")
option(AUDIOCPP_USE_SYSTEM_OPENSSL
    "Use system OpenSSL for native model management instead of bundled BoringSSL"
    OFF)
if (AUDIOCPP_USE_SYSTEM_OPENSSL AND NOT AUDIOCPP_BUILD_NATIVE_MODEL_MANAGER)
    message(FATAL_ERROR
        "AUDIOCPP_USE_SYSTEM_OPENSSL requires AUDIOCPP_BUILD_NATIVE_MODEL_MANAGER=ON")
endif()
option(ENGINE_ENABLE_CPU_ALL_VARIANTS
    "Build CPU backends as dynamic libraries with per-ISA variants (for Docker/portable builds)"
    OFF)
option(ENGINE_BUILD_EXAMPLES "Build example binaries" OFF)
option(ENGINE_BUILD_TESTS "Build framework unit tests" OFF)
option(ENGINE_BUILD_EXTENDED_TESTS "Build extended non-model tests and probes" OFF)
option(ENGINE_BUILD_MODEL_TESTS "Build model-specific tests and probes" OFF)
option(ENGINE_BUILD_WARMBENCH "Build warmbench helper binaries" OFF)
option(AUDIOCPP_BUILD_C_API "Build the C ABI (include/audiocpp.h) as a shared library" OFF)
set(AUDIOCPP_C_API_MODEL_ROOT "" CACHE PATH
    "Models root for the C API model test. Unset leaves the test registered but skipping.")
set(AUDIOCPP_RKNN_TEST_ENGINE "" CACHE PATH
    "A .rknn engine compiled for the build host's SoC, for the RKNN sidecar test. Unset leaves the test registered but skipping.")
set(AUDIOCPP_GGML_SOURCE_DIR
    "${CMAKE_CURRENT_SOURCE_DIR}/external/ggml"
    CACHE PATH "ggml source tree to build with AudioCPP")

set(AUDIOCPP_GIT_SHA "unknown")
set(AUDIOCPP_GIT_DATE "unknown")
find_package(Git QUIET)
if (GIT_FOUND)
    execute_process(
        COMMAND ${GIT_EXECUTABLE} rev-parse --short HEAD
        WORKING_DIRECTORY "${CMAKE_CURRENT_SOURCE_DIR}"
        OUTPUT_VARIABLE AUDIOCPP_GIT_SHA
        OUTPUT_STRIP_TRAILING_WHITESPACE
        ERROR_QUIET)
    execute_process(
        COMMAND ${GIT_EXECUTABLE} show -s --format=%cs HEAD
        WORKING_DIRECTORY "${CMAKE_CURRENT_SOURCE_DIR}"
        OUTPUT_VARIABLE AUDIOCPP_GIT_DATE
        OUTPUT_STRIP_TRAILING_WHITESPACE
        ERROR_QUIET)
endif()

set(AUDIOCPP_BUILD_BACKENDS "cpu")
if (ENGINE_ENABLE_CUDA)
    string(APPEND AUDIOCPP_BUILD_BACKENDS ",cuda")
endif()
if (ENGINE_ENABLE_HIP)
    string(APPEND AUDIOCPP_BUILD_BACKENDS ",hip")
endif()
if (ENGINE_ENABLE_VULKAN)
    string(APPEND AUDIOCPP_BUILD_BACKENDS ",vulkan")
endif()
if (ENGINE_ENABLE_METAL)
    string(APPEND AUDIOCPP_BUILD_BACKENDS ",metal")
endif()
configure_file(
    "${CMAKE_CURRENT_SOURCE_DIR}/app/common/build_info_config.h.in"
    "${CMAKE_CURRENT_BINARY_DIR}/generated/build_info_config.h"
    @ONLY
)

# HIP compiles ggml's CUDA backend sources as HIP code, so the two options drive the
# same ggml backend and cannot coexist: enabling both starts CUDA backend setup and
# HIP backend setup in the same configure, which fails in confusing ways downstream.
if (ENGINE_ENABLE_CUDA AND ENGINE_ENABLE_HIP)
    message(FATAL_ERROR
        "ENGINE_ENABLE_CUDA and ENGINE_ENABLE_HIP are mutually exclusive; "
        "configure with exactly one of them (e.g. -DENGINE_ENABLE_CUDA=OFF -DENGINE_ENABLE_HIP=ON)")
endif()

set(AUDIOCPP_HIP_STRIX_HALO_OPTIMIZATIONS_ACTIVE OFF)
if (ENGINE_HIP_STRIX_HALO_OPTIMIZATIONS)
    if (NOT ENGINE_ENABLE_HIP)
        message(FATAL_ERROR
            "ENGINE_HIP_STRIX_HALO_OPTIMIZATIONS=ON requires ENGINE_ENABLE_HIP=ON")
    endif()

    # Match ggml HIP target precedence: an explicit CMake HIP architecture wins,
    # followed by GPU_TARGETS and then the legacy AMDGPU_TARGETS spelling.
    if (CMAKE_HIP_ARCHITECTURES)
        set(AUDIOCPP_EFFECTIVE_HIP_TARGETS ${CMAKE_HIP_ARCHITECTURES})
        set(AUDIOCPP_EFFECTIVE_HIP_TARGET_SOURCE "CMAKE_HIP_ARCHITECTURES")
    elseif (GPU_TARGETS)
        set(AUDIOCPP_EFFECTIVE_HIP_TARGETS ${GPU_TARGETS})
        set(AUDIOCPP_EFFECTIVE_HIP_TARGET_SOURCE "GPU_TARGETS")
    elseif (AMDGPU_TARGETS)
        set(AUDIOCPP_EFFECTIVE_HIP_TARGETS ${AMDGPU_TARGETS})
        set(AUDIOCPP_EFFECTIVE_HIP_TARGET_SOURCE "AMDGPU_TARGETS")
    else()
        message(FATAL_ERROR
            "ENGINE_HIP_STRIX_HALO_OPTIMIZATIONS=ON requires exactly one explicit HIP target, gfx1151; "
            "set -DGPU_TARGETS=gfx1151 (no project-wide gfx1151 default is provided)")
    endif()

    list(LENGTH AUDIOCPP_EFFECTIVE_HIP_TARGETS AUDIOCPP_EFFECTIVE_HIP_TARGET_COUNT)
    if (NOT AUDIOCPP_EFFECTIVE_HIP_TARGET_COUNT EQUAL 1 OR
        NOT "${AUDIOCPP_EFFECTIVE_HIP_TARGETS}" STREQUAL "gfx1151")
        message(FATAL_ERROR
            "ENGINE_HIP_STRIX_HALO_OPTIMIZATIONS=ON requires the effective HIP target list to be exactly "
            "gfx1151; ${AUDIOCPP_EFFECTIVE_HIP_TARGET_SOURCE}='${AUDIOCPP_EFFECTIVE_HIP_TARGETS}'")
    endif()

    # hipBLASLt regressed decoder compute by roughly 3 seconds on gfx1151.
    # Keep the validated retained-graph path on rocBLAS; the generic HIP build
    # remains unchanged because this option is both target- and backend-gated.
    set(GGML_HIP_HIPBLASLT OFF CACHE BOOL
        "Use hipBLASLt instead of hipBLAS (rocBLAS) for HIP GEMM" FORCE)
    set(AUDIOCPP_HIP_STRIX_HALO_OPTIMIZATIONS_ACTIVE ON)
    message(STATUS "Strix Halo HIP optimizations enabled for gfx1151 with rocBLAS")
endif()

if (ENGINE_ENABLE_CUDA AND NOT ENGINE_ENABLE_HIP)
    # Require at least CUDA 12.0. Older toolkits configure fine but die at the
    # first .cu file. Run this before enable_language(CUDA) because the version
    # is needed for the default arch list below.
    find_package(CUDAToolkit 12.0 REQUIRED)

    # Define default CUDA architectures for every build path.
    # We stay close to llama.cpp/ggml-cuda here, unless there is a reason
    # to deviate. See llama.cpp's ggml-cuda CMakeLists.txt for additional
    # comments.
    # Must be set before enable_language(CUDA), as that call otherwise seeds
    # CMAKE_CUDA_ARCHITECTURES from nvcc's default arch (sm_75 for CUDA 13,
    # sm_52 for CUDA 12), silently making every build without an explicit
    # list single-arch (the seed ignores the local GPU).
    # This can be overridden with -DCMAKE_CUDA_ARCHITECTURES=... or CUDAARCHS;
    # set to "native" to build for the local GPU only (needs CMake >= 3.24).

    # "native" fallback for CMake < 3.24
    if ((CMAKE_CUDA_ARCHITECTURES STREQUAL "native" OR "$ENV{CUDAARCHS}" STREQUAL "native")
        AND CMAKE_VERSION VERSION_LESS "3.24")
        message(STATUS "CMAKE_CUDA_ARCHITECTURES=native requires CMake >= 3.24; "
                       "falling back to the portable default architecture list")
        set(CMAKE_CUDA_ARCHITECTURES "")
    endif()

    # Define default archs
    if ("${CMAKE_CUDA_ARCHITECTURES}" STREQUAL "" AND "$ENV{CUDAARCHS}" STREQUAL "")
        # Remembered so the summary below can tell the difference between "the
        # portable list is what you asked for" and "the portable list is what you
        # got by not asking".
        set(_AUDIOCPP_CUDA_ARCH_DEFAULTED TRUE)
        # XX-virtual = PTX (JIT, forward-compatible); XX-real = SASS (native).
        if (CUDAToolkit_VERSION VERSION_LESS "13")
            # 50/61/70 (Maxwell/Pascal/Volta) unsupported by CUDA 13 toolkits
            list(APPEND CMAKE_CUDA_ARCHITECTURES 50-virtual 61-virtual 70-virtual)
        endif()
        list(APPEND CMAKE_CUDA_ARCHITECTURES 75-real 80-virtual 86-real)
        if (CUDAToolkit_VERSION VERSION_GREATER_EQUAL "11.8")
            list(APPEND CMAKE_CUDA_ARCHITECTURES 89-real 90-virtual)
        endif()
        if (CUDAToolkit_VERSION VERSION_GREATER_EQUAL "12.8")
            # 12Xa, not 12X: Blackwell FP4 tensor cores are arch-specific.
            list(APPEND CMAKE_CUDA_ARCHITECTURES 120a-real)
        endif()
        if (CUDAToolkit_VERSION VERSION_GREATER_EQUAL "12.9")
            list(APPEND CMAKE_CUDA_ARCHITECTURES 121a-real)
        endif()
    endif()

    enable_language(CUDA)

    # Upgrade plain 12X (user value or resolved "native") to 12Xa (Blackwell).
    foreach (_ARCHS_VAR IN ITEMS CMAKE_CUDA_ARCHITECTURES CMAKE_CUDA_ARCHITECTURES_NATIVE)
        set(_FIXED_ARCHS "")
        foreach (_ARCH IN LISTS ${_ARCHS_VAR})
            if (_ARCH MATCHES "^12[0-9](-real|-virtual)?$")
                string(REGEX REPLACE "^(12[0-9])((-real|-virtual)?)$" "\\1a\\2" _FIXED_ARCH "${_ARCH}")
                message(STATUS "Replacing ${_ARCH} in ${_ARCHS_VAR} with ${_FIXED_ARCH}")
                list(APPEND _FIXED_ARCHS "${_FIXED_ARCH}")
            else()
                list(APPEND _FIXED_ARCHS "${_ARCH}")
            endif()
        endforeach()
        set(${_ARCHS_VAR} "${_FIXED_ARCHS}")
    endforeach()

    # Resolve "native" for the log; left as-is when no GPU was detected
    # (nvcc then warns and uses its default arch).
    if (CMAKE_CUDA_ARCHITECTURES STREQUAL "native" AND CMAKE_CUDA_ARCHITECTURES_NATIVE MATCHES "^[0-9]+(a|f)?(-real|-virtual)?(;[0-9]+(a|f)?(-real|-virtual)?|;)*$")
        set(CMAKE_CUDA_ARCHITECTURES "${CMAKE_CUDA_ARCHITECTURES_NATIVE}")
    endif()
    message(STATUS "Using CMAKE_CUDA_ARCHITECTURES=${CMAKE_CUDA_ARCHITECTURES} CMAKE_CUDA_ARCHITECTURES_NATIVE=${CMAKE_CUDA_ARCHITECTURES_NATIVE}")

    # The line above prints both halves of this and interprets neither: how many
    # architectures are about to be compiled, and which one this host can run.
    # The option is documented (README, docs/build/linux.md), but configure is
    # where the choice is actually being made, so say it here too.
    if (_AUDIOCPP_CUDA_ARCH_DEFAULTED AND
        CMAKE_CUDA_ARCHITECTURES_NATIVE MATCHES "^[0-9]")
        list(LENGTH CMAKE_CUDA_ARCHITECTURES _AUDIOCPP_ARCH_COUNT)
        message(STATUS "  portable default: ${_AUDIOCPP_ARCH_COUNT} architectures, so every .cu file is compiled ${_AUDIOCPP_ARCH_COUNT} times")
        message(STATUS "  this host is ${CMAKE_CUDA_ARCHITECTURES_NATIVE}; for a local build roughly 3x faster:")
        message(STATUS "      -DCMAKE_CUDA_ARCHITECTURES=native    (the result then runs on this GPU only)")
        message(STATUS "      scripts/build_linux.sh: --cuda-arch native")
    endif()

    if (NOT MSVC)
        set(CMAKE_CUDA_FLAGS_DEBUG "${AUDIOCPP_DEBUG_OPT_FLAGS}" CACHE STRING "Optimized debug CUDA flags" FORCE)
    endif()
endif()

# ⚠ Outside the CUDA block on purpose: ccache is not a CUDA concern, and a
# CPU-only, Vulkan or HIP build benefits from it just as much. GGML_CCACHE is
# forced OFF above, so ggml's own probe never runs and nothing caches anything
# unless the caller wires a launcher up by hand.
find_program(AUDIOCPP_CCACHE_PROGRAM ccache)
if (AUDIOCPP_CCACHE_PROGRAM AND
    NOT CMAKE_C_COMPILER_LAUNCHER AND NOT CMAKE_CXX_COMPILER_LAUNCHER)
    message(STATUS "ccache is installed (${AUDIOCPP_CCACHE_PROGRAM}) and NOT in use. Rebuilds are much faster with:")
    message(STATUS "    -DCMAKE_C_COMPILER_LAUNCHER=ccache -DCMAKE_CXX_COMPILER_LAUNCHER=ccache -DCMAKE_CUDA_COMPILER_LAUNCHER=ccache")
    message(STATUS "    scripts/build_linux.sh: --ccache")
endif()

if (ENGINE_ENABLE_OPENMP)
    find_package(OpenMP REQUIRED COMPONENTS CXX)
endif()

if (AUDIOCPP_BUILD_C_API)
    # engine_runtime and ggml are static/object libraries here; linking them into
    # a shared libaudiocpp requires position-independent code throughout.
    set(CMAKE_POSITION_INDEPENDENT_CODE ON)
endif()

if (ENGINE_ENABLE_CPU_ALL_VARIANTS)
    message(STATUS "CPU all variants mode: enabling shared libraries + backends as dynamic libs")
    set(BUILD_SHARED_LIBS ON CACHE BOOL "Build shared libraries" FORCE)
    set(GGML_BACKEND_DL ON CACHE BOOL "Build backends as dynamic libraries" FORCE)
    set(GGML_CPU_ALL_VARIANTS ON CACHE BOOL "Build all CPU backend variants" FORCE)
    set(GGML_NATIVE OFF CACHE BOOL "Build ggml CPU kernels with native host ISA flags" FORCE)
    # Ensure libraries in the same directory as the binary are found at runtime
    set(CMAKE_INSTALL_RPATH "\$ORIGIN")
    set(CMAKE_BUILD_WITH_INSTALL_RPATH ON)
else()
    set(BUILD_SHARED_LIBS OFF CACHE BOOL "Build shared libraries" FORCE)
    set(GGML_BACKEND_DL OFF CACHE BOOL "Build backends as dynamic libraries" FORCE)
    set(GGML_CPU_ALL_VARIANTS OFF CACHE BOOL "Build all CPU backend variants" FORCE)
    set(GGML_NATIVE ${ENGINE_ENABLE_NATIVE_CPU} CACHE BOOL "Build ggml CPU kernels with native host ISA flags" FORCE)
endif()

set(GGML_BUILD_EXAMPLES OFF CACHE BOOL "Do not build upstream examples" FORCE)
set(GGML_BUILD_TESTS OFF CACHE BOOL "Do not build upstream tests" FORCE)
set(GGML_ALL_WARNINGS OFF CACHE BOOL "Keep upstream warnings quiet in this sample" FORCE)
set(GGML_CCACHE OFF CACHE BOOL "Do not probe for ccache in this local setup" FORCE)
set(GGML_CUDA ${ENGINE_ENABLE_CUDA} CACHE BOOL "Build ggml with CUDA backend support" FORCE)
set(GGML_HIP ${ENGINE_ENABLE_HIP} CACHE BOOL "Build ggml with HIP backend support" FORCE)
set(GGML_VULKAN ${ENGINE_ENABLE_VULKAN} CACHE BOOL "Build ggml with Vulkan backend support" FORCE)
set(GGML_METAL ${ENGINE_ENABLE_METAL} CACHE BOOL "Build ggml with Metal backend support" FORCE)
set(GGML_LLAMAFILE ${ENGINE_ENABLE_LLAMAFILE} CACHE BOOL "Build ggml with llamafile SGEMM support" FORCE)
set(GGML_CUDA_GRAPHS ${ENGINE_ENABLE_CUDA_GRAPHS} CACHE BOOL "Enable ggml CUDA graphs support" FORCE)
# HIP graphs are controlled by a separate upstream option that defaults to ON;
# keep it in sync so ENGINE_ENABLE_CUDA_GRAPHS=OFF also disables graphs on HIP builds
set(GGML_HIP_GRAPHS ${ENGINE_ENABLE_CUDA_GRAPHS} CACHE BOOL "Enable ggml HIP graphs support" FORCE)

set(SPM_ENABLE_SHARED OFF CACHE BOOL "Build sentencepiece shared libs" FORCE)
set(SPM_BUILD_TEST OFF CACHE BOOL "Do not build sentencepiece tests" FORCE)
set(SPM_ENABLE_TCMALLOC OFF CACHE BOOL "Do not probe for tcmalloc" FORCE)

if (NOT EXISTS "${AUDIOCPP_GGML_SOURCE_DIR}/CMakeLists.txt")
    message(FATAL_ERROR "AUDIOCPP_GGML_SOURCE_DIR does not point to a ggml source tree: ${AUDIOCPP_GGML_SOURCE_DIR}")
endif()
message(STATUS "Using ggml source: ${AUDIOCPP_GGML_SOURCE_DIR}")
add_subdirectory("${AUDIOCPP_GGML_SOURCE_DIR}" "${CMAKE_CURRENT_BINARY_DIR}/ggml")
add_subdirectory(external/sentencepiece EXCLUDE_FROM_ALL)

add_compile_options(
    $<$<COMPILE_LANG_AND_ID:CXX,GNU,Clang>:-Wall>
    $<$<COMPILE_LANG_AND_ID:CXX,GNU,Clang>:-Wextra>
    $<$<COMPILE_LANG_AND_ID:CXX,GNU,Clang>:-Wpedantic>
    $<$<COMPILE_LANG_AND_ID:CXX,GNU,Clang>:-pedantic-errors>
    $<$<COMPILE_LANG_AND_ID:CXX,MSVC>:/permissive->
)

set(AUDIOCPP_MODEL_SET "full" CACHE STRING "Model composite to build: full, core, or custom")
set_property(CACHE AUDIOCPP_MODEL_SET PROPERTY STRINGS full core custom)
set(AUDIOCPP_MODELS "" CACHE STRING "Comma or semicolon separated model targets for AUDIOCPP_MODEL_SET=custom")
set(AUDIOCPP_LIBYAML_COMPILE_DEFINITIONS
    _POSIX_C_SOURCE=200809L
    YAML_DECLARE_STATIC=1
    YAML_VERSION_MAJOR=0
    YAML_VERSION_MINOR=2
    YAML_VERSION_PATCH=5
    YAML_VERSION_STRING=\"0.2.5\"
)

function(audiocpp_configure_runtime_object target_name)
    target_include_directories(${target_name} PUBLIC
        ${CMAKE_CURRENT_SOURCE_DIR}/include
    )
    target_include_directories(${target_name} PRIVATE
        ${CMAKE_CURRENT_BINARY_DIR}/generated
        ${CMAKE_CURRENT_SOURCE_DIR}/external/sentencepiece/src
        ${CMAKE_CURRENT_SOURCE_DIR}/external/llama_tokenizer
        ${CMAKE_CURRENT_SOURCE_DIR}/external/cJSON
        ${CMAKE_CURRENT_SOURCE_DIR}/external/libyaml/include
    )
    target_compile_definitions(${target_name} PRIVATE ${AUDIOCPP_LIBYAML_COMPILE_DEFINITIONS})
    if (AUDIOCPP_HIP_STRIX_HALO_OPTIMIZATIONS_ACTIVE)
        target_compile_definitions(${target_name} PRIVATE ENGINE_HIP_STRIX_HALO_OPTIMIZATIONS=1)
    endif()
    if (ENGINE_ENABLE_RKNN)
        target_compile_definitions(${target_name} PRIVATE ENGINE_HAVE_RKNN=1)
        if (ENGINE_RKNN_INCLUDE_DIR)
            target_include_directories(${target_name} PRIVATE ${ENGINE_RKNN_INCLUDE_DIR})
        endif()
    endif()
    target_link_libraries(${target_name} PRIVATE ggml sentencepiece)
    if (ENGINE_ENABLE_OPENMP)
        target_link_libraries(${target_name} PRIVATE OpenMP::OpenMP_CXX)
        if (MSVC)
            target_compile_options(${target_name} PRIVATE $<$<COMPILE_LANGUAGE:CXX>:/openmp:experimental>)
        endif()
    endif()
endfunction()

function(audiocpp_add_model target_name)
    cmake_parse_arguments(ARG "" "" "SOURCES;INCLUDES;LOADERS;ALIASES;DEPENDS;OBJECT_LIBRARIES" ${ARGN})
    add_library(engine_model_${target_name} OBJECT ${ARG_SOURCES})
    set_target_properties(engine_model_${target_name} PROPERTIES EXCLUDE_FROM_ALL TRUE)
    audiocpp_configure_runtime_object(engine_model_${target_name})
    if (ARG_OBJECT_LIBRARIES)
        target_link_libraries(engine_model_${target_name} PRIVATE ${ARG_OBJECT_LIBRARIES})
    endif()
    if (ENGINE_ENABLE_NEON)
        target_compile_definitions(engine_model_${target_name} PRIVATE ENGINE_HAVE_NEON=1)
    endif()
    set(AUDIOCPP_MODEL_TARGETS ${AUDIOCPP_MODEL_TARGETS} ${target_name} PARENT_SCOPE)
    set(AUDIOCPP_MODEL_${target_name}_INCLUDES "${ARG_INCLUDES}" PARENT_SCOPE)
    set(AUDIOCPP_MODEL_${target_name}_LOADERS "${ARG_LOADERS}" PARENT_SCOPE)
    set(AUDIOCPP_MODEL_${target_name}_DEPENDS "${ARG_DEPENDS}" PARENT_SCOPE)
    set(AUDIOCPP_MODEL_${target_name}_OBJECT_LIBRARIES "${ARG_OBJECT_LIBRARIES}" PARENT_SCOPE)
    set(AUDIOCPP_MODEL_ALIAS_${target_name} ${target_name} PARENT_SCOPE)
    foreach(alias IN LISTS ARG_ALIASES)
        set(AUDIOCPP_MODEL_ALIAS_${alias} ${target_name} PARENT_SCOPE)
    endforeach()
endfunction()

if (ENGINE_ENABLE_NEON)
    add_library(engine_neon OBJECT
        src/framework/neon/tensor.cpp
        src/framework/neon/operators.cpp
        src/framework/neon/tensor_operators.cpp
        src/framework/neon/execution.cpp
        src/framework/neon/worker_team.cpp
        src/framework/neon/blocks.cpp
        src/framework/neon/hifigan.cpp
        src/framework/neon/hift.cpp
        src/framework/neon/vocoder_operators.cpp
        src/framework/neon/kernels.cpp
    )
    target_include_directories(engine_neon PUBLIC "${CMAKE_CURRENT_SOURCE_DIR}/include")
    target_compile_features(engine_neon PUBLIC cxx_std_17)
    target_link_libraries(engine_neon PUBLIC Threads::Threads)
    target_compile_definitions(engine_neon PUBLIC ENGINE_HAVE_NEON=1)
endif()

add_library(engine_core OBJECT
    src/framework/assets/resource_bundle.cpp
    src/framework/model_spec/package.cpp
    src/framework/model_spec/options.cpp
    src/framework/model_spec/schema.cpp
    src/framework/assets/tensor_source.cpp
    src/framework/assets/weight_norm_folding.cpp
    src/framework/assets/weight_adapter.cpp
    src/framework/assets/weight_adapter_source.cpp
    src/framework/assets/torch_bin.cpp
    src/framework/core/module.cpp
    src/framework/core/backend.cpp
    src/framework/core/attention_fallback.cpp
    src/framework/core/deferred_tensor_writer.cpp
    src/framework/core/disk_expert_weight_store.cpp
    src/framework/core/execution_context.cpp
    src/framework/core/host_memory.cpp
    src/framework/core/load_progress.cpp
    src/framework/runtime/model.cpp
    src/framework/model_spec/metadata.cpp
    src/framework/runtime/session.cpp
    src/framework/runtime/task_vocabulary.cpp
    src/framework/runtime/partial_text.cpp
    src/framework/runtime/artifacts.cpp
    src/framework/runtime/cache.cpp
    src/framework/runtime/graph_executor.cpp
    src/framework/runtime/graph_optimizer.cpp
    src/framework/runtime/host_ops.cpp
    src/framework/runtime/kv_cache.cpp
    src/framework/runtime/bounded_static_kv_decode.cpp
    src/framework/runtime/greedy_causal_decoder.cpp
    src/framework/runtime/options.cpp
    src/framework/runtime/session_base.cpp
    src/framework/runtime/workspace.cpp
    src/framework/midi/midi_file.cpp
    src/framework/io/filesystem.cpp
    src/framework/text/espeak_phonemizer.cpp
    src/framework/text/espeak_data.cpp
    src/framework/io/config.cpp
    src/framework/io/json.cpp
    src/framework/io/text.cpp
    src/framework/text/chinese_normalization.cpp
    src/framework/text/chinese_variant.cpp
    src/framework/text/chunking.cpp
    src/framework/text/subtitle.cpp
    src/framework/text/moss_tts_normalization.cpp
    src/framework/text/text_normalization.cpp
    src/framework/text/unicode_normalization.cpp
    src/framework/io/yaml.cpp
    src/framework/io/binary.cpp
    src/framework/io/safetensors.cpp
    src/framework/debug/trace.cpp
    src/framework/debug/profiler.cpp
    src/framework/rknn/engine.cpp
    src/framework/rknn/engine_resources.cpp
    src/framework/decoders/moss_tts_delay/backbone.cpp
    src/framework/decoders/moss_tts_delay/config.cpp
    src/framework/decoders/moss_tts_delay/delay_decoder.cpp
    src/framework/decoders/moss_tts_delay/heads.cpp
    src/framework/decoders/moss_tts_delay/prompt.cpp
    src/framework/decoders/tdt_decoder_runner.cpp
    src/framework/decoders/tdt_decoder_greedy_customized.cpp
    src/framework/decoders/tdt_decoder_greedy_duration_loop.cpp
    src/framework/decoders/tdt_decoder_greedy_separate_heads.cpp
    src/framework/modules/linear_module.cpp
    src/framework/modules/packed_linear_weights.cpp
    src/framework/modules/convnext_modules.cpp
    src/framework/modules/vocoders/vocos_vocoder.cpp
    src/framework/modules/primitive_modules.cpp
    src/framework/modules/activation_modules.cpp
    src/framework/modules/norm_modules.cpp
    src/framework/modules/lookup_modules.cpp
    src/framework/modules/multi_codebook_embedding.cpp
    src/framework/modules/structural_modules.cpp
    src/framework/modules/asr_helpers.cpp
    src/framework/modules/conv_modules.cpp
    src/framework/modules/zipformer_modules.cpp
    src/framework/modules/speech_encoders/whisper_embedding.cpp
    src/framework/modules/speech_decoders/ming_flow_runtime.cpp
    src/framework/modules/speech_encoders/whisper_frontend.cpp
    src/framework/modules/speaker_encoders/campplus_encoder.cpp
    src/framework/modules/speech_encoders/s3_tokenizer.cpp
    src/framework/codecs/nemo_nano_codec_runtime.cpp
    src/framework/codecs/wan_video_vae_runtime.cpp
    src/framework/modules/pitch_extractors/rmvpe_pitch_extractor.cpp
    src/framework/modules/vocoders/bigvgan_vocoder.cpp
    src/framework/modules/vocoders/hifigan_vocoder.cpp
    src/framework/modules/vocoders/hift_vocoder.cpp
    src/framework/modules/speech_encoders/wav2vec2_encoder.cpp
    src/framework/modules/speech_encoders/wav2vec2_bert_encoder.cpp
    src/framework/modules/speech_encoders/wavlm_encoder.cpp
    src/framework/modules/speech_encoders/sanm.cpp
    src/framework/modules/optimizations/fast_conv_modules.cpp
    src/framework/modules/optimizations/fast_projection_modules.cpp
    src/framework/modules/recurrent_modules.cpp
    src/framework/modules/streaming_conv_modules.cpp
    src/framework/modules/attention/positional_encoding.cpp
    src/framework/modules/feed_forward_modules.cpp
    src/framework/modules/attention/self_attention.cpp
    src/framework/modules/attention/longformer_attention.cpp
    src/framework/modules/attention/longformer_attention_gpu.cpp
    src/framework/modules/attention/common_relative_attention.cpp
    src/framework/modules/attention/relative_attention.cpp
    src/framework/modules/attention/scaled_dot_product_attention.cpp
    src/framework/modules/attention/grouped_query_attention.cpp
    src/framework/modules/attention/projected_grouped_self_attention.cpp
    src/framework/modules/attention/streaming_frame_attention_cache.cpp
    src/framework/modules/transformers/decoder.cpp
    src/framework/modules/transformers/causal_decoder.cpp
    src/framework/modules/transformers/causal_decoder_runtime.cpp
    src/framework/modules/transformers/qwen35_decoder_runtime.cpp
    src/framework/modules/text_encoders/qwen3_vl_encoder_runtime.cpp
    src/framework/modules/speech_encoders/qwen3_audio_encoder_runtime.cpp
    src/framework/modules/attention/cross_attention.cpp
    src/framework/modules/transformers/transformer_blocks.cpp
    src/framework/modules/conditioning_modules.cpp
    src/framework/modules/conformer_modules.cpp
    src/framework/modules/ebranchformer_modules.cpp
    src/framework/sampling/flow_sampler_runtime.cpp
    src/framework/modules/predictor_modules.cpp
    src/framework/modules/positional_modules.cpp
    src/framework/modules/text_encoders/t5_base_encoder.cpp
    src/framework/modules/text_encoders/t5_gemma_encoder.cpp
    src/framework/modules/transformers/gemma_decoder.cpp
    src/framework/modules/optimizations/fast_kv_modules.cpp
    src/framework/tokenizers/hf_tokenizer_json.cpp
    src/framework/tokenizers/llama_bpe.cpp
    src/framework/tokenizers/qwen_bpe_bundle.cpp
    external/llama_tokenizer/bpe-core.cpp
    external/llama_tokenizer/unicode.cpp
    external/llama_tokenizer/unicode-data.cpp
    src/framework/tokenizers/sentencepiece.cpp
    src/framework/audio/fft.cpp
    src/framework/audio/kaldi_fbank.cpp
    src/framework/audio/activity.cpp
    src/framework/audio/probability_segmenter.cpp
    src/framework/audio/activity_graph.cpp
    src/framework/audio/chunking.cpp
    src/framework/audio/conversion.cpp
    src/framework/audio/utilities/deepfilternet2.cpp
    src/framework/audio/dsp.cpp
    src/framework/audio/nemo_mel_frontend.cpp
    src/framework/audio/mel_spectrogram_frontend.cpp
    src/framework/audio/reference_audio_frontend.cpp
    src/framework/audio/utilities/flashsr.cpp
    src/framework/audio/utilities/gtcrn.cpp
    src/framework/audio/istft_graph.cpp
    src/framework/audio/mixing.cpp
    src/framework/audio/resampling.cpp
    src/framework/audio/output.cpp
    src/framework/audio/utilities/rnnoise.cpp
    src/framework/audio/utilities/utility_api.cpp
    src/framework/audio/waveform_ops.cpp
    src/framework/audio/wav_reader.cpp
    src/framework/audio/wav_writer.cpp
    src/framework/audio/utilities/zipenhancer.cpp
    src/framework/runtime/flow_kv_cache.cpp
    src/framework/modules/speaker_encoders/ecapa_tdnn_runtime.cpp
    src/framework/modules/speaker_encoders/titanet_runtime.cpp
    src/framework/codecs/fsq_audio_codec_runtime.cpp
    src/framework/codecs/fish_dac_codec_runtime.cpp
    src/framework/codecs/mel_latent_vae44k_runtime.cpp
    src/framework/codecs/mimi_codec_runtime.cpp
    src/framework/codecs/moss_audio_tokenizer_codec_runtime.cpp
    src/framework/codecs/neural_audio.cpp
    src/framework/codecs/oobleck_audio_vae_runtime.cpp
    src/framework/codecs/redae_codec_runtime.cpp
    src/framework/codecs/snac_decoder_runtime.cpp
    src/framework/codecs/ming_audio_vae_runtime.cpp
    src/framework/codecs/s3gen_runtime.cpp
    src/framework/conditioners/clap_audio_conditioner_runtime.cpp
    src/framework/conditioners/cav_mae_st_conditioner_runtime.cpp
    src/framework/conditioners/musicgen_style_conditioner_runtime.cpp
    src/framework/conditioners/open_clip_conditioner_runtime.cpp
    src/framework/conditioners/synchformer_conditioner_runtime.cpp
    src/models/controlfoley/flow_denoiser.cpp
    src/framework/sampling/greedy_decode.cpp
    src/framework/sampling/diffusion_math.cpp
    src/framework/sampling/hf_sampler.cpp
    src/framework/sampling/noise.cpp
    src/framework/sampling/torch_random.cpp
)
audiocpp_configure_runtime_object(engine_core)

# These two small built-in model paths are kept outside the selectable composite list for now.
add_library(engine_model_silero_vad OBJECT
    src/models/silero_vad/assets.cpp
    src/models/silero_vad/runtime.cpp
    src/models/silero_vad/session.cpp
)
set_target_properties(engine_model_silero_vad PROPERTIES EXCLUDE_FROM_ALL TRUE)
audiocpp_configure_runtime_object(engine_model_silero_vad)

add_library(engine_model_marblenet_vad OBJECT
    src/models/marblenet_vad/assets.cpp
    src/models/marblenet_vad/runtime.cpp
    src/models/marblenet_vad/session.cpp
)
set_target_properties(engine_model_marblenet_vad PROPERTIES EXCLUDE_FROM_ALL TRUE)
audiocpp_configure_runtime_object(engine_model_marblenet_vad)

audiocpp_add_model(moss_transcribe_diarize
    SOURCES
        src/models/moss_transcribe_diarize/runtime.cpp
        src/models/moss_transcribe_diarize/session.cpp
    INCLUDES
        engine/models/moss_transcribe_diarize/session.h
    LOADERS
        engine::models::moss_transcribe_diarize::make_moss_transcribe_diarize_loader
)

audiocpp_add_model(universr
    SOURCES
        src/models/universr/assets.cpp
        src/models/universr/network.cpp
        src/models/universr/runtime.cpp
        src/models/universr/frontend.cpp
        src/models/universr/session.cpp
    INCLUDES
        engine/models/universr/session.h
    LOADERS
        engine::models::universr::make_universr_loader
)

audiocpp_add_model(sam_audio
    SOURCES
        src/models/sam_audio/video.cpp
        src/models/sam_audio/frontend.cpp
        src/models/sam_audio/vision.cpp
        src/models/sam_audio/codec.cpp
        src/models/sam_audio/conditioner.cpp
        src/models/sam_audio/dit.cpp
        src/models/sam_audio/session.cpp
    INCLUDES
        engine/models/sam_audio/session.h
    LOADERS
        engine::models::sam_audio::make_sam_audio_loader
)

audiocpp_add_model(mossformer2
    SOURCES
        src/models/mossformer2/assets.cpp
        src/models/mossformer2/runtime.cpp
        src/models/mossformer2/session.cpp
    INCLUDES
        engine/models/mossformer2/session.h
    LOADERS
        engine::models::mossformer2::make_mossformer2_loader
)

audiocpp_add_model(tf_gridnet
    SOURCES
        src/models/tf_gridnet/assets.cpp
        src/models/tf_gridnet/runtime.cpp
        src/models/tf_gridnet/session.cpp
    INCLUDES
        engine/models/tf_gridnet/session.h
    LOADERS
        engine::models::tf_gridnet::make_tf_gridnet_loader
)

audiocpp_add_model(apollo
    SOURCES
        src/models/apollo/assets.cpp
        src/models/apollo/runtime.cpp
        src/models/apollo/session.cpp
    INCLUDES
        engine/models/apollo/session.h
    LOADERS
        engine::models::apollo::make_apollo_loader
)

audiocpp_add_model(sidon
    SOURCES
        src/models/sidon/runtime.cpp
        src/models/sidon/frontend.cpp
        src/models/sidon/session.cpp
    INCLUDES
        engine/models/sidon/session.h
    LOADERS
        engine::models::sidon::make_sidon_loader
)

audiocpp_add_model(tone_color_vc
    SOURCES
        src/models/tone_color_vc/runtime.cpp
        src/models/tone_color_vc/session.cpp
    INCLUDES
        engine/models/tone_color_vc/session.h
    LOADERS
        engine::models::tone_color_vc::make_tone_color_vc_loader
)

audiocpp_add_model(firered_vad
    SOURCES
        src/models/firered_vad/runtime.cpp
        src/models/firered_vad/session.cpp
    INCLUDES
        engine/models/firered_vad/session.h
    LOADERS
        engine::models::firered_vad::make_firered_vad_loader
)

audiocpp_add_model(pulsevad
    SOURCES
        src/models/pulsevad/runtime.cpp
        src/models/pulsevad/session.cpp
    INCLUDES
        engine/models/pulsevad/session.h
    LOADERS
        engine::models::pulsevad::make_pulsevad_loader
)

audiocpp_add_model(smart_turn
    SOURCES
        src/models/smart_turn/runtime.cpp
        src/models/smart_turn/session.cpp
    INCLUDES
        engine/models/smart_turn/session.h
    LOADERS
        engine::models::smart_turn::make_smart_turn_loader
)

audiocpp_add_model(ced
    SOURCES
        src/models/ced/assets.cpp
        src/models/ced/frontend.cpp
        src/models/ced/runtime.cpp
        src/models/ced/session.cpp
    INCLUDES
        engine/models/ced/session.h
    LOADERS
        engine::models::ced::make_ced_loader
)

audiocpp_add_model(ast_audioset
    SOURCES
        src/models/ast_audioset/assets.cpp
        src/models/ast_audioset/frontend.cpp
        src/models/ast_audioset/runtime.cpp
        src/models/ast_audioset/session.cpp
    INCLUDES
        engine/models/ast_audioset/session.h
    LOADERS
        engine::models::ast_audioset::make_ast_audioset_loader
)

audiocpp_add_model(roformer
    SOURCES
        src/models/roformer/assets.cpp
        src/models/roformer/runtime.cpp
        src/models/roformer/session.cpp
    INCLUDES
        engine/models/roformer/session.h
    LOADERS
        engine::models::roformer::make_mel_band_roformer_loader
        engine::models::roformer::make_bs_roformer_loader
    ALIASES
        mel_band_roformer
        bs_roformer
)

audiocpp_add_model(demucs
    SOURCES
        src/models/demucs/assets.cpp
        src/models/demucs/frontend.cpp
        src/models/demucs/pipeline.cpp
        src/models/demucs/postprocess.cpp
        src/models/demucs/session.cpp
    INCLUDES
        engine/models/demucs/session.h
    LOADERS
        engine::models::demucs::make_htdemucs_loader
        engine::models::demucs::make_htdemucs_6stems_loader
    ALIASES
        htdemucs
        htdemucs_6stems
        htdemucs_6s
)

audiocpp_add_model(kokoro_tts
    SOURCES
        src/models/kokoro_tts/assets.cpp
        src/models/kokoro_tts/g2p_multilingual.cpp
        src/models/kokoro_tts/frontend.cpp
        src/models/kokoro_tts/predictor.cpp
        src/models/kokoro_tts/decoder.cpp
        src/models/kokoro_tts/rknn_generator.cpp
        src/models/kokoro_tts/plbert.cpp
        src/models/kokoro_tts/session.cpp
    INCLUDES engine/models/kokoro_tts/session.h
    LOADERS engine::models::kokoro_tts::make_kokoro_tts_loader
)
if(MSVC)
    target_compile_options(engine_model_kokoro_tts PRIVATE /utf-8)
endif()
if(ENGINE_ENABLE_NEON)
    target_sources(engine_model_kokoro_tts PRIVATE src/models/kokoro_tts/neon_generator.cpp)
endif()
if(AUDIOCPP_STATIC_ESPEAK)
    target_compile_definitions(engine_model_kokoro_tts PRIVATE AUDIOCPP_STATIC_ESPEAK=1)
endif()

audiocpp_add_model(omnivoice
    SOURCES
        src/models/omnivoice/assets.cpp
        src/models/omnivoice/tokenizer_text.cpp
        src/models/omnivoice/audio_tokenizer.cpp
        src/models/omnivoice/prompt_builder.cpp
        src/models/omnivoice/generator.cpp
        src/models/omnivoice/postprocess.cpp
        src/models/omnivoice/session.cpp
        src/models/omnivoice/loader.cpp
    INCLUDES
        engine/models/omnivoice/loader.h
    LOADERS
        engine::models::omnivoice::make_omnivoice_loader
)

audiocpp_add_model(reuse
    SOURCES
        src/community_models/reuse/frontend.cpp
        src/community_models/reuse/runtime.cpp
        src/community_models/reuse/session.cpp
    INCLUDES
        engine/community_models/reuse/session.h
    LOADERS
        engine::models::reuse::make_reuse_loader
)

audiocpp_add_model(glm_tts
    SOURCES
        src/community_models/glm_tts/assets.cpp
        src/community_models/glm_tts/flow.cpp
        src/community_models/glm_tts/frontend.cpp
        src/community_models/glm_tts/llama.cpp
        src/community_models/glm_tts/session.cpp
        src/community_models/glm_tts/prompt.cpp
        src/community_models/glm_tts/speech_tokenizer.cpp
        src/community_models/glm_tts/tokenizer_text.cpp
    INCLUDES
        engine/community_models/glm_tts/session.h
    LOADERS
        engine::models::glm_tts::make_glm_tts_loader
    DEPENDS
        outetts
)

audiocpp_add_model(kroko_asr
    SOURCES
        src/community_models/kroko_asr/assets.cpp
        src/community_models/kroko_asr/decoder.cpp
        src/community_models/kroko_asr/encoder.cpp
        src/community_models/kroko_asr/frontend.cpp
        src/community_models/kroko_asr/session.cpp
        src/community_models/kroko_asr/tokenizer.cpp
        src/community_models/kroko_asr/zipformer.cpp
    INCLUDES
        engine/community_models/kroko_asr/session.h
    LOADERS
        engine::models::kroko_asr::make_kroko_asr_loader
)

audiocpp_add_model(open_wake_word
    SOURCES
        src/models/open_wake_word/frontend.cpp
        src/models/open_wake_word/weights.cpp
        src/models/open_wake_word/runtime.cpp
        src/models/open_wake_word/session.cpp
    INCLUDES
        engine/models/open_wake_word/session.h
    LOADERS
        engine::models::open_wake_word::make_open_wake_word_loader
)
if (ENGINE_ENABLE_NEON)
    target_sources(engine_model_open_wake_word PRIVATE src/models/open_wake_word/neon_embedding.cpp)
endif()

audiocpp_add_model(micro_wake_word
    SOURCES
        src/models/micro_wake_word/frontend.cpp
        src/models/micro_wake_word/runtime.cpp
        src/models/micro_wake_word/session.cpp
    INCLUDES
        engine/models/micro_wake_word/session.h
    LOADERS
        engine::models::micro_wake_word::make_micro_wake_word_loader
)
if(ENGINE_ENABLE_NEON)
    target_sources(engine_model_micro_wake_word PRIVATE src/models/micro_wake_word/neon_network.cpp)
endif()

audiocpp_add_model(sherpa_kws
    SOURCES
        src/models/sherpa_kws/assets.cpp
        src/models/sherpa_kws/encoder.cpp
        src/models/sherpa_kws/frontend.cpp
        src/models/sherpa_kws/keyword_decoder.cpp
        src/models/sherpa_kws/session.cpp
        src/models/sherpa_kws/transducer.cpp
        src/models/sherpa_kws/zipformer.cpp
    INCLUDES
        engine/models/sherpa_kws/session.h
    LOADERS
        engine::models::sherpa_kws::make_sherpa_kws_loader
)

audiocpp_add_model(outetts
    SOURCES
        src/community_models/outetts/assets.cpp
        src/community_models/outetts/dac.cpp
        src/community_models/outetts/llama.cpp
        src/community_models/outetts/session.cpp
        src/community_models/outetts/tokenizer.cpp
    INCLUDES
        engine/community_models/outetts/session.h
    LOADERS
        engine::models::outetts::make_outetts_loader
    DEPENDS
        qwen3_asr
        qwen3_forced_aligner
)

audiocpp_add_model(voxtral_tts
    SOURCES
        src/models/voxtral_tts/assets.cpp
        src/models/voxtral_tts/ar.cpp
        src/models/voxtral_tts/acoustic.cpp
        src/models/voxtral_tts/codec.cpp
        src/models/voxtral_tts/session.cpp
    INCLUDES
        engine/models/voxtral_tts/session.h
    LOADERS
        engine::models::voxtral_tts::make_voxtral_tts_loader
)

audiocpp_add_model(kugelaudio
    SOURCES
        src/models/kugelaudio/ar.cpp
        src/models/kugelaudio/assets.cpp
        src/models/kugelaudio/codec.cpp
        src/models/kugelaudio/diffusion.cpp
        src/models/kugelaudio/generator.cpp
        src/models/kugelaudio/scheduler.cpp
        src/models/kugelaudio/session.cpp
    INCLUDES
        engine/models/kugelaudio/session.h
    LOADERS
        engine::models::kugelaudio::make_kugelaudio_loader
)

audiocpp_add_model(maya1
    SOURCES
        src/models/maya1/assets.cpp
        src/models/maya1/generator.cpp
        src/models/maya1/session.cpp
        src/models/maya1/tokenizer.cpp
    INCLUDES
        engine/models/maya1/session.h
    LOADERS
        engine::models::maya1::make_maya1_loader
)

audiocpp_add_model(parakeet_tdt
    SOURCES
        src/community_models/parakeet_tdt/assets.cpp
        src/community_models/parakeet_tdt/session.cpp
        src/community_models/parakeet_tdt/encoder.cpp
        src/community_models/parakeet_tdt/decoder.cpp
        src/community_models/parakeet_tdt/hotwords.cpp
        src/community_models/parakeet_tdt/phonon_streaming.cpp
        src/community_models/parakeet_tdt/frontend.cpp
        src/community_models/parakeet_tdt/weights.cpp
    INCLUDES
        engine/community_models/parakeet_tdt/session.h
    LOADERS
        engine::community_models::parakeet_tdt::make_parakeet_tdt_loader
)

audiocpp_add_model(personaplex
    SOURCES
        src/models/personaplex/assets.cpp
        src/models/personaplex/depformer.cpp
        src/models/personaplex/lm_runtime.cpp
        src/models/personaplex/request.cpp
        src/models/personaplex/session.cpp
    INCLUDES
        engine/models/personaplex/session.h
    LOADERS
        engine::models::personaplex::make_personaplex_loader
)

set(AUDIOCPP_SANOTTS_NEON_SOURCES src/community_models/sanotts/neon_decoder.cpp)
if (ENGINE_ENABLE_NEON)
    set(AUDIOCPP_SANOTTS_NEON_SOURCES
        src/community_models/sanotts/neon_library_decoder.cpp
        src/community_models/sanotts/neon_weights.cpp)
endif()
audiocpp_add_model(sanotts
    SOURCES
        src/community_models/sanotts/assets.cpp
        src/community_models/sanotts/frontend.cpp
        ${AUDIOCPP_SANOTTS_NEON_SOURCES}
        src/community_models/sanotts/piper_runtime.cpp
        src/community_models/sanotts/runtime.cpp
        src/community_models/sanotts/session.cpp
    INCLUDES
        engine/community_models/sanotts/session.h
    LOADERS
        engine::models::sanotts::make_sanotts_loader
)

audiocpp_add_model(inflect_v2
    SOURCES
        src/community_models/inflect_v2/assets.cpp
        src/community_models/inflect_v2/frontend.cpp
        src/community_models/inflect_v2/runtime.cpp
        src/community_models/inflect_v2/rknn_decoder.cpp
        src/community_models/inflect_v2/session.cpp
    INCLUDES
        engine/community_models/inflect_v2/session.h
    LOADERS
        engine::models::inflect_v2::make_inflect_v2_loader
)

audiocpp_add_model(piper_tts
    SOURCES
        src/community_models/piper_tts/assets.cpp
        src/community_models/piper_tts/frontend.cpp
        src/community_models/piper_tts/vits_runtime.cpp
        src/community_models/piper_tts/rknn_decoder.cpp
        src/community_models/piper_tts/rknn_duration_encoder.cpp
        src/community_models/piper_tts/session.cpp
    INCLUDES
        engine/community_models/piper_tts/session.h
    LOADERS
        engine::models::piper_tts::make_piper_tts_loader
)

audiocpp_add_model(kitten_tts2
    SOURCES
        src/community_models/kitten_tts2/session.cpp
        src/community_models/kitten_tts2/decoder.cpp
        src/community_models/kitten_tts2/speaker.cpp
    INCLUDES engine/community_models/kitten_tts2/session.h
    LOADERS engine::community_models::kitten_tts2::make_kitten_tts2_loader
    DEPENDS chatterbox)

if (ENGINE_ENABLE_NEON)
    target_sources(engine_model_kitten_tts2 PRIVATE src/community_models/kitten_tts2/neon_vocoder.cpp)
endif()

audiocpp_add_model(kitten_tts
    SOURCES
        src/community_models/kitten_tts/assets.cpp
        src/community_models/kitten_tts/decoder.cpp
        src/community_models/kitten_tts/frontend.cpp
        src/community_models/kitten_tts/plbert.cpp
        src/community_models/kitten_tts/predictor.cpp
        src/community_models/kitten_tts/session.cpp
    INCLUDES
        engine/community_models/kitten_tts/session.h
    LOADERS
        engine::models::kitten_tts::make_kitten_tts_loader
)
if(MSVC)
    target_compile_options(engine_model_kitten_tts PRIVATE /utf-8)
endif()

audiocpp_add_model(audio_flamingo
    SOURCES
        src/models/audio_flamingo/assets.cpp
        src/models/audio_flamingo/audio_encoder.cpp
        src/models/audio_flamingo/decoder.cpp
        src/models/audio_flamingo/frontend.cpp
        src/models/audio_flamingo/projector.cpp
        src/models/audio_flamingo/session.cpp
        src/models/audio_flamingo/tokenizer_text.cpp
    INCLUDES
        engine/models/audio_flamingo/session.h
    LOADERS
        engine::models::audio_flamingo::make_audio_flamingo_loader
)

audiocpp_add_model(samsone
    SOURCES
        src/models/samsone/assets.cpp
        src/models/samsone/runtime.cpp
        src/models/samsone/session.cpp
    INCLUDES
        engine/models/samsone/session.h
    LOADERS
        engine::models::samsone::make_samsone_loader
)

audiocpp_add_model(f5_tts
    SOURCES
        src/community_models/f5_tts/session.cpp
        src/community_models/f5_tts/runtime.cpp
        src/community_models/f5_tts/weights.cpp
        src/community_models/f5_tts/dit_modules.cpp
        src/community_models/f5_tts/synthesize.cpp
    INCLUDES
        engine/community_models/f5_tts/session.h
        engine/community_models/f5_tts/runtime.h
    LOADERS
        engine::models::f5_tts::make_f5_tts_loader
    ALIASES
        habibi
        habibi_tts
)

audiocpp_add_model(zipvoice
    SOURCES
        src/community_models/zipvoice/session.cpp
        src/community_models/zipvoice/synthesize.cpp
        src/community_models/zipvoice/emilia_tokenizer.cpp
        src/community_models/zipvoice/jieba_segmenter.cpp
        src/community_models/zipvoice/weights.cpp
        src/community_models/zipvoice/zipformer.cpp
    INCLUDES
        engine/community_models/zipvoice/session.h
        engine/community_models/zipvoice/synthesize.h
    LOADERS
        engine::models::zipvoice::make_zipvoice_loader
)

audiocpp_add_model(minimax_h3
    SOURCES
        src/community_models/minimax_h3/assets.cpp
        src/community_models/minimax_h3/audio_vae_decoder.cpp
        src/community_models/minimax_h3/dit_acceleration.cpp
        src/community_models/minimax_h3/dit_denoiser.cpp
        src/community_models/minimax_h3/generation_plan.cpp
        src/community_models/minimax_h3/pipeline.cpp
        src/community_models/minimax_h3/sampler.cpp
        src/community_models/minimax_h3/session.cpp
        src/community_models/minimax_h3/video_vae_decoder.cpp
    INCLUDES
        engine/community_models/minimax_h3/session.h
    LOADERS
        engine::models::minimax_h3::make_minimax_h3_loader
)

audiocpp_add_model(latentsync
    SOURCES
        src/community_models/latentsync/whisper.cpp
        src/community_models/latentsync/unet.cpp
        src/community_models/latentsync/sd_vae.cpp
        src/community_models/latentsync/face.cpp
        src/community_models/latentsync/sampling.cpp
        src/community_models/latentsync/pipeline.cpp
        src/community_models/latentsync/video.cpp
        src/community_models/latentsync/session.cpp
    INCLUDES
        engine/community_models/latentsync/session.h
    LOADERS
        engine::models::latentsync::make_latentsync_loader
)
if (GGML_CUDA_CUDNN_CONV2D)
    target_compile_definitions(engine_model_latentsync PRIVATE GGML_CUDA_CUDNN_CONV2D=1)
endif()

audiocpp_add_model(liveavatar
    SOURCES
        src/community_models/liveavatar/assets.cpp
        src/community_models/liveavatar/pipeline.cpp
        src/community_models/liveavatar/pipeline_state.cpp
        src/community_models/liveavatar/pipeline_text_encoder.cpp
        src/community_models/liveavatar/pipeline_vae.cpp
        src/community_models/liveavatar/pipeline_denoiser.cpp
        src/community_models/liveavatar/pipeline_liveavatar.cpp
        src/community_models/liveavatar/pipeline_offline.cpp
        src/community_models/liveavatar/wan_s2v_audio_conditioner.cpp
        src/community_models/liveavatar/session.cpp
    INCLUDES
        engine/community_models/liveavatar/session.h
    LOADERS
        engine::community_models::liveavatar::make_liveavatar_loader
)

audiocpp_add_model(minimax_music3
    SOURCES
        src/community_models/minimax_music3/assets.cpp
        src/community_models/minimax_music3/prompt.cpp
        src/community_models/minimax_music3/global_lm.cpp
        src/community_models/minimax_music3/depth_decoder.cpp
        src/community_models/minimax_music3/ar_runtime.cpp
        src/community_models/minimax_music3/condition_encoder.cpp
        src/community_models/minimax_music3/flow_transformer.cpp
        src/community_models/minimax_music3/flow_sampler.cpp
        src/community_models/minimax_music3/vocoder.cpp
        src/community_models/minimax_music3/pipeline.cpp
        src/community_models/minimax_music3/session.cpp
    INCLUDES
        engine/community_models/minimax_music3/session.h
    LOADERS
        engine::models::minimax_music3::make_minimax_music3_loader
)

audiocpp_add_model(neutts
    SOURCES
        src/models/neutts/ar.cpp
        src/models/neutts/assets.cpp
        src/models/neutts/backbone.cpp
        src/models/neutts/codec.cpp
        src/models/neutts/prompt.cpp
        src/models/neutts/session.cpp
    INCLUDES
        engine/models/neutts/ar.h
        engine/models/neutts/assets.h
        engine/models/neutts/backbone.h
        engine/models/neutts/codec.h
        engine/models/neutts/prompt.h
        engine/models/neutts/session.h
    LOADERS
        engine::models::neutts::make_neutts_loader
)

audiocpp_add_model(echo_tts
    SOURCES
        src/community_models/echo_tts/session.cpp
        src/community_models/echo_tts/tokenizer.cpp
        src/community_models/echo_tts/latent_post.cpp
        src/community_models/echo_tts/sampler.cpp
        src/community_models/echo_tts/dit.cpp
    INCLUDES
        engine/community_models/echo_tts/session.h
        engine/community_models/echo_tts/config.h
        engine/community_models/echo_tts/tokenizer.h
        engine/community_models/echo_tts/latent_post.h
        engine/community_models/echo_tts/sampler.h
        engine/community_models/echo_tts/dit.h
    LOADERS
        engine::models::echo_tts::make_echo_tts_loader
)

if (MSVC)
    set_source_files_properties(
        src/community_models/inflect_v2/frontend.cpp
        PROPERTIES COMPILE_OPTIONS /utf-8
    )
endif()

audiocpp_add_model(pocket_tts
    SOURCES
        src/models/pocket_tts/assets.cpp
        src/models/pocket_tts/backend_weights.cpp
        src/models/pocket_tts/acoustic_model.cpp
        src/models/pocket_tts/audio_decoder.cpp
        src/models/pocket_tts/flow_lm.cpp
        src/models/pocket_tts/mimi_decoder.cpp
        src/models/pocket_tts/mimi_encoder.cpp
        src/models/pocket_tts/text_conditioner.cpp
        src/models/pocket_tts/voice_state_assets.cpp
        src/models/pocket_tts/voice_conditioner.cpp
        src/models/pocket_tts/loader.cpp
        src/models/pocket_tts/session.cpp
        src/models/pocket_tts/gradium_phonon.cpp
    INCLUDES
        engine/models/pocket_tts/loader.h
    LOADERS
        engine::models::pocket_tts::make_pocket_tts_loader
        engine::models::pocket_tts::make_gradium_phonon_loader
    ALIASES
        gradium_phonon
)

audiocpp_add_model(miocodec
    SOURCES
        src/models/miocodec/assets.cpp
        src/models/miocodec/weights.cpp
        src/models/miocodec/audio_pipeline.cpp
        src/models/miocodec/graph_ops.cpp
        src/models/miocodec/graphs.cpp
        src/models/miocodec/session.cpp
        src/models/miocodec/loader.cpp
    INCLUDES
        engine/models/miocodec/loader.h
    LOADERS
        engine::models::miocodec::make_miocodec_loader
)

audiocpp_add_model(miotts
    SOURCES
        src/models/miotts/assets.cpp
        src/models/miotts/tokenizer.cpp
        src/models/miotts/causal_lm.cpp
        src/models/miotts/session.cpp
        src/models/miotts/loader.cpp
    INCLUDES
        engine/models/miotts/loader.h
    LOADERS
        engine::models::miotts::make_miotts_loader
    DEPENDS
        miocodec
        qwen3_asr
)

audiocpp_add_model(moss
    SOURCES
        src/models/moss/moss_tts_local/assets.cpp
        src/models/moss/moss_tts_local/backbone.cpp
        src/models/moss/moss_tts_nano/assets.cpp
        src/models/moss/moss_tts_nano/generator.cpp
        src/models/moss/moss_tts_nano/global_transformer.cpp
        src/models/moss/moss_tts_nano/local_frame_decoder.cpp
        src/models/moss/moss_tts_nano/loader.cpp
        src/models/moss/moss_tts_nano/prompt_builder.cpp
        src/models/moss/moss_tts_nano/session.cpp
        src/models/moss/moss_tts_nano/tokenizer_text.cpp
        src/models/moss/moss_tts_local/depth_transformer.cpp
        src/models/moss/moss_tts_local/generator.cpp
        src/models/moss/moss_tts_local/loader.cpp
        src/models/moss/moss_tts_local/tokenizer_text.cpp
        src/models/moss/moss_tts_local/session.cpp
    INCLUDES
        engine/models/moss/moss_tts_local/loader.h
        engine/models/moss/moss_tts_nano/loader.h
    LOADERS
        engine::models::moss_tts_local::make_moss_tts_local_loader
        engine::models::moss_tts_nano::make_moss_tts_nano_loader
    ALIASES
        moss_tts_local
        moss_tts_nano
)

audiocpp_add_model(moss_ttsd
    SOURCES
        src/community_models/moss_ttsd/assets.cpp
        src/community_models/moss_ttsd/prompt.cpp
        src/community_models/moss_ttsd/session.cpp
        src/community_models/moss_ttsd/loader.cpp
    INCLUDES
        engine/community_models/moss_ttsd/loader.h
    LOADERS
        engine::models::moss_ttsd::make_moss_ttsd_loader
)

audiocpp_add_model(moss_tts_v15
    SOURCES
        src/community_models/moss_tts_v15/assets.cpp
        src/community_models/moss_tts_v15/prompt.cpp
        src/community_models/moss_tts_v15/session.cpp
        src/community_models/moss_tts_v15/loader.cpp
    INCLUDES
        engine/community_models/moss_tts_v15/loader.h
    LOADERS
        engine::models::moss_tts_v15::make_moss_tts_v15_loader
)

audiocpp_add_model(moss_voicegen
    SOURCES
        src/community_models/moss_voicegen/assets.cpp
        src/community_models/moss_voicegen/loader.cpp
        src/community_models/moss_voicegen/session.cpp
        src/community_models/moss_voicegen/tokenizer_text.cpp
    INCLUDES
        engine/community_models/moss_voicegen/loader.h
    LOADERS
        engine::models::moss_voicegen::make_moss_voicegen_loader
    DEPENDS
        moss
)

audiocpp_add_model(muscriptor
    SOURCES
        src/models/muscriptor/assets.cpp
        src/models/muscriptor/frontend.cpp
        src/models/muscriptor/tokenizer.cpp
        src/models/muscriptor/decoder.cpp
        src/models/muscriptor/session.cpp
    INCLUDES
        engine/models/muscriptor/session.h
    LOADERS
        engine::models::muscriptor::make_muscriptor_loader
)

audiocpp_add_model(voxcpm2
    SOURCES
        src/models/voxcpm2/assets.cpp
        src/models/voxcpm2/audiovae.cpp
        src/models/voxcpm2/generator.cpp
        src/models/voxcpm2/loader.cpp
        src/models/voxcpm2/minicpm.cpp
        src/models/voxcpm2/session.cpp
        src/models/voxcpm2/tokenizer_text.cpp
    INCLUDES
        engine/models/voxcpm2/loader.h
    LOADERS
        engine::models::voxcpm2::make_voxcpm2_loader
)

audiocpp_add_model(voxcpm1
    SOURCES
        src/community_models/voxcpm1/assets.cpp
        src/community_models/voxcpm1/audiovae.cpp
        src/community_models/voxcpm1/generator.cpp
        src/community_models/voxcpm1/minicpm.cpp
        src/community_models/voxcpm1/session.cpp
        src/community_models/voxcpm1/tokenizer_text.cpp
    INCLUDES
        engine/community_models/voxcpm1/session.h
    LOADERS
        engine::community_models::voxcpm1::make_voxcpm1_loader
)

audiocpp_add_model(vibevoice
    SOURCES
        src/models/vibevoice/assets.cpp
        src/models/vibevoice/connector.cpp
        src/models/vibevoice/decoder.cpp
        src/models/vibevoice/diffusion_head.cpp
        src/models/vibevoice/diffusion_sampler.cpp
        src/models/vibevoice/generator.cpp
        src/models/vibevoice/loader.cpp
        src/models/vibevoice/lora.cpp
        src/models/vibevoice/scheduler.cpp
        src/models/vibevoice/session.cpp
        src/models/vibevoice/tokenizer_audio.cpp
        src/models/vibevoice/tokenizer_text.cpp
    INCLUDES
        engine/models/vibevoice/loader.h
    LOADERS
        engine::models::vibevoice::make_vibevoice_loader
)

audiocpp_add_model(vibevoice_asr
    SOURCES
        src/models/vibevoice_asr/assets.cpp
        src/models/vibevoice_asr/connector.cpp
        src/models/vibevoice_asr/frontend.cpp
        src/models/vibevoice_asr/loader.cpp
        src/models/vibevoice_asr/postprocess.cpp
        src/models/vibevoice_asr/session.cpp
        src/models/vibevoice_asr/speech_encoder.cpp
        src/models/vibevoice_asr/speech_tokenizer.cpp
        src/models/vibevoice_asr/text_decoder.cpp
        src/models/vibevoice_asr/tokenizer_text.cpp
    INCLUDES
        engine/models/vibevoice_asr/loader.h
    LOADERS
        engine::models::vibevoice_asr::make_vibevoice_asr_loader
        engine::models::vibevoice_asr::make_vibevoice_asr_streaming_loader
)

audiocpp_add_model(voxtral_realtime
    SOURCES
        src/models/voxtral_realtime/assets.cpp
        src/models/voxtral_realtime/tokenizer_text.cpp
        src/models/voxtral_realtime/frontend.cpp
        src/models/voxtral_realtime/audio_encoder.cpp
        src/models/voxtral_realtime/text_decoder.cpp
        src/models/voxtral_realtime/session.cpp
        src/models/voxtral_realtime/loader.cpp
    INCLUDES
        engine/models/voxtral_realtime/loader.h
    LOADERS
        engine::models::voxtral_realtime::make_voxtral_realtime_loader
)

audiocpp_add_model(fish_audio
    SOURCES
        src/models/fish_audio/ar.cpp
        src/models/fish_audio/assets.cpp
        src/models/fish_audio/generator.cpp
        src/models/fish_audio/loader.cpp
        src/models/fish_audio/prompt_builder.cpp
        src/models/fish_audio/session.cpp
        src/models/fish_audio/tokenizer_text.cpp
    INCLUDES
        engine/models/fish_audio/loader.h
    LOADERS
        engine::models::fish_audio::make_fish_audio_loader
)

add_library(utf8proc_vendor OBJECT external/utf8proc/utf8proc.c)
set_target_properties(utf8proc_vendor PROPERTIES EXCLUDE_FROM_ALL TRUE POSITION_INDEPENDENT_CODE ON)
target_include_directories(utf8proc_vendor PUBLIC external/utf8proc)
target_compile_definitions(utf8proc_vendor PUBLIC UTF8PROC_STATIC)

audiocpp_add_model(ming3_audio
    SOURCES
        src/models/ming3_audio/assets.cpp
        src/models/ming3_audio/audio_encoder.cpp
        src/models/ming3_audio/backbone.cpp
        src/models/ming3_audio/frontend.cpp
        src/models/ming3_audio/session.cpp
        src/models/ming3_audio/talker.cpp
        src/models/ming3_audio/talker_frontend.cpp
    INCLUDES
        engine/models/ming3_audio/session.h
    LOADERS
        engine::models::ming3_audio::make_ming3_audio_loader
)

audiocpp_add_model(ming3_tts
    SOURCES
        src/models/ming3_tts/assets.cpp
        src/models/ming3_tts/generator.cpp
        src/models/ming3_tts/prompt.cpp
        src/models/ming3_tts/session.cpp
    INCLUDES
        engine/models/ming3_tts/session.h
    LOADERS
        engine::models::ming3_tts::make_ming3_tts_loader
)

audiocpp_add_model(plapre
    SOURCES
        src/models/plapre/assets.cpp
        src/models/plapre/frontend.cpp
        src/models/plapre/generator.cpp
        src/models/plapre/kanade.cpp
        src/models/plapre/prompt.cpp
        src/models/plapre/session.cpp
    INCLUDES
        engine/models/plapre/session.h
    LOADERS
        engine::models::plapre::make_plapre_loader
    OBJECT_LIBRARIES
        utf8proc_vendor
)

audiocpp_add_model(fish_drama3
    SOURCES
        src/community_models/fish_drama3/assets.cpp
        src/community_models/fish_drama3/backbone.cpp
        src/community_models/fish_drama3/generator.cpp
        src/community_models/fish_drama3/loader.cpp
        src/community_models/fish_drama3/prompt.cpp
        src/community_models/fish_drama3/session.cpp
        src/community_models/fish_drama3/tokenizer_text.cpp
        src/community_models/fish_drama3/tower.cpp
    INCLUDES
        engine/community_models/fish_drama3/loader.h
    LOADERS
        engine::models::fish_drama3::make_fish_drama3_loader
    OBJECT_LIBRARIES
        utf8proc_vendor
)

audiocpp_add_model(audio8_tts
    SOURCES
        src/community_models/audio8_tts/ar.cpp
        src/community_models/audio8_tts/assets.cpp
        src/community_models/audio8_tts/codec.cpp
        src/community_models/audio8_tts/generator.cpp
        src/community_models/audio8_tts/prompt_builder.cpp
        src/community_models/audio8_tts/session.cpp
        src/community_models/audio8_tts/tokenizer_text.cpp
    INCLUDES
        engine/community_models/audio8_tts/session.h
    LOADERS
        engine::models::audio8_tts::make_audio8_tts_loader
)

audiocpp_add_model(magpie_tts
    SOURCES
        src/models/magpie_tts/assets.cpp
        src/models/magpie_tts/request.cpp
        src/models/magpie_tts/runtime.cpp
        src/models/magpie_tts/session.cpp
        src/models/magpie_tts/tokenizer_text.cpp
    INCLUDES
        engine/models/magpie_tts/request.h
        engine/models/magpie_tts/runtime.h
        engine/models/magpie_tts/session.h
        engine/models/magpie_tts/tokenizer_text.h
        engine/models/magpie_tts/types.h
    LOADERS
        engine::models::magpie_tts::make_magpie_tts_loader
)

audiocpp_add_model(confucius4_tts
    SOURCES
        src/models/confucius4_tts/assets.cpp
        src/models/confucius4_tts/audio_features.cpp
        src/models/confucius4_tts/request.cpp
        src/models/confucius4_tts/s2a.cpp
        src/models/confucius4_tts/semantic_encoder.cpp
        src/models/confucius4_tts/session.cpp
        src/models/confucius4_tts/style_encoder.cpp
        src/models/confucius4_tts/t2s.cpp
        src/models/confucius4_tts/tokenizer_text.cpp
        src/models/confucius4_tts/vocoder.cpp
    INCLUDES
        engine/models/confucius4_tts/session.h
    LOADERS
        engine::models::confucius4_tts::make_confucius4_tts_loader
)

audiocpp_add_model(dots_tts
    SOURCES
        src/models/dots_tts/assets.cpp
        src/models/dots_tts/audio_features.cpp
        src/models/dots_tts/audio_vae.cpp
        src/models/dots_tts/flow.cpp
        src/models/dots_tts/flow_meanflow.cpp
        src/models/dots_tts/flow_soar.cpp
        src/models/dots_tts/latent.cpp
        src/models/dots_tts/llm.cpp
        src/models/dots_tts/patch_encoder.cpp
        src/models/dots_tts/request.cpp
        src/models/dots_tts/session.cpp
        src/models/dots_tts/tokenizer.cpp
    INCLUDES
        engine/models/dots_tts/session.h
    LOADERS
        engine::models::dots_tts::make_dots_tts_loader
)

audiocpp_add_model(dramabox
    SOURCES
        src/models/dramabox/assets.cpp
        src/models/dramabox/audio_vae.cpp
        src/models/dramabox/dit.cpp
        src/models/dramabox/gemma_tokenizer.cpp
        src/models/dramabox/gemma3_encoder.cpp
        src/models/dramabox/latent_state.cpp
        src/models/dramabox/prompt_connector.cpp
        src/models/dramabox/session.cpp
        src/models/dramabox/vocoder.cpp
    INCLUDES
        engine/models/dramabox/session.h
    LOADERS
        engine::models::dramabox::make_dramabox_loader
)

audiocpp_add_model(fun_asr_nano
    SOURCES
        src/models/fun_asr_nano/assets.cpp
        src/models/fun_asr_nano/frontend.cpp
        src/models/fun_asr_nano/encoder.cpp
        src/models/fun_asr_nano/adaptor.cpp
        src/models/fun_asr_nano/tokenizer_text.cpp
        src/models/fun_asr_nano/prompt.cpp
        src/models/fun_asr_nano/decoder.cpp
        src/models/fun_asr_nano/session.cpp
        src/models/fun_asr_nano/loader.cpp
    INCLUDES
        engine/models/fun_asr_nano/loader.h
    LOADERS
        engine::models::fun_asr_nano::make_fun_asr_nano_loader
)

audiocpp_add_model(sense_asr
    SOURCES
        src/community_models/sense_asr/assets.cpp
        src/community_models/sense_asr/frontend.cpp
        src/community_models/sense_asr/encoder.cpp
        src/community_models/sense_asr/session.cpp
    INCLUDES
        engine/community_models/sense_asr/session.h
    LOADERS
        engine::community_models::sense_asr::make_sense_asr_loader
)

audiocpp_add_model(heartmula
    SOURCES
        src/models/heartmula/assets.cpp
        src/models/heartmula/codec.cpp
        src/models/heartmula/generator.cpp
        src/models/heartmula/mula.cpp
        src/models/heartmula/session.cpp
        src/models/heartmula/tokenizer_text.cpp
    INCLUDES
        engine/models/heartmula/session.h
    LOADERS
        engine::models::heartmula::make_heartmula_loader
)

audiocpp_add_model(higgs_audio_stt
    SOURCES
        src/models/higgs_audio_stt/assets.cpp
        src/models/higgs_audio_stt/tokenizer_text.cpp
        src/models/higgs_audio_stt/frontend_whisper.cpp
        src/models/higgs_audio_stt/audio_encoder.cpp
        src/models/higgs_audio_stt/text_decoder.cpp
        src/models/higgs_audio_stt/prompt_asr.cpp
        src/models/higgs_audio_stt/postprocess.cpp
        src/models/higgs_audio_stt/session.cpp
    INCLUDES
        engine/models/higgs_audio_stt/session.h
    LOADERS
        engine::models::higgs_audio_stt::make_higgs_audio_stt_loader
)

audiocpp_add_model(higgs_audio_tts
    SOURCES
        src/models/higgs_audio_tts/ar.cpp
        src/models/higgs_audio_tts/assets.cpp
        src/models/higgs_audio_tts/codec.cpp
        src/models/higgs_audio_tts/codebooks.cpp
        src/models/higgs_audio_tts/generator.cpp
        src/models/higgs_audio_tts/loader.cpp
        src/models/higgs_audio_tts/sampler.cpp
        src/models/higgs_audio_tts/session.cpp
        src/models/higgs_audio_tts/tokenizer_text.cpp
    INCLUDES
        engine/models/higgs_audio_tts/loader.h
    LOADERS
        engine::models::higgs_audio_tts::make_higgs_audio_tts_loader
)

audiocpp_add_model(hviske_asr
    SOURCES
        src/models/hviske_asr/assets.cpp
        src/models/hviske_asr/frontend.cpp
        src/models/hviske_asr/encoder.cpp
        src/models/hviske_asr/decoder.cpp
        src/models/hviske_asr/session.cpp
        src/models/hviske_asr/weights.cpp
    INCLUDES
        engine/models/hviske_asr/session.h
    LOADERS
        engine::models::hviske_asr::make_hviske_asr_loader
)

audiocpp_add_model(hviske_asr_v6
    SOURCES
        src/models/hviske_asr_v6/assets.cpp
        src/models/hviske_asr_v6/frontend.cpp
        src/models/hviske_asr_v6/encoder.cpp
        src/models/hviske_asr_v6/decoder.cpp
        src/models/hviske_asr_v6/session.cpp
    INCLUDES
        engine/models/hviske_asr_v6/session.h
    LOADERS
        engine::models::hviske_asr_v6::make_hviske_asr_v6_loader
)

audiocpp_add_model(irodori_tts
    SOURCES
        src/models/irodori_tts/assets.cpp
        src/models/irodori_tts/codec.cpp
        src/models/irodori_tts/condition_encoder.cpp
        src/models/irodori_tts/duration.cpp
        src/models/irodori_tts/rf_dit.cpp
        src/models/irodori_tts/session.cpp
        src/models/irodori_tts/tokenizer_text.cpp
    INCLUDES
        engine/models/irodori_tts/session.h
    LOADERS
        engine::models::irodori_tts::make_irodori_tts_loader
)

audiocpp_add_model(index_tts2
    SOURCES
        src/models/index_tts2/assets.cpp
        src/models/index_tts2/audio_features.cpp
        src/models/index_tts2/gpt.cpp
        src/models/index_tts2/loader.cpp
        src/models/index_tts2/qwen_emotion.cpp
        src/models/index_tts2/request.cpp
        src/models/index_tts2/s2mel.cpp
        src/models/index_tts2/semantic_codec.cpp
        src/models/index_tts2/semantic_encoder.cpp
        src/models/index_tts2/session.cpp
        src/models/index_tts2/style_encoder.cpp
        src/models/index_tts2/tokenizer_text.cpp
        src/models/index_tts2/vocoder.cpp
    INCLUDES
        engine/models/index_tts2/loader.h
    LOADERS
        engine::models::index_tts2::make_index_tts2_loader
)

audiocpp_add_model(nemotron_asr
    SOURCES
        src/models/nemotron_asr/assets.cpp
        src/models/nemotron_asr/frontend.cpp
        src/models/nemotron_asr/encoder.cpp
        src/models/nemotron_asr/decoder.cpp
        src/models/nemotron_asr/session.cpp
        src/models/nemotron_asr/speaker_tagging.cpp
        src/models/nemotron_asr/masked_asr.cpp
        src/models/nemotron_asr/loader.cpp
        src/models/nemotron_asr/weights.cpp
    INCLUDES
        engine/models/nemotron_asr/loader.h
    LOADERS
        engine::models::nemotron_asr::make_nemotron_asr_loader
)

audiocpp_add_model(gigaam_asr
    SOURCES
        src/models/gigaam_asr/assets.cpp
        src/models/gigaam_asr/conformer.cpp
        src/models/gigaam_asr/frontend.cpp
        src/models/gigaam_asr/weights.cpp
        src/models/gigaam_asr/runtime.cpp
        src/models/gigaam_asr/session.cpp
    INCLUDES
        engine/models/gigaam_asr/model.h
    LOADERS
        engine::models::gigaam_asr::make_gigaam_asr_loader
)

audiocpp_add_model(crisperwhisper
    SOURCES
        src/models/crisperwhisper/assets.cpp
        src/models/crisperwhisper/alignment.cpp
        src/models/crisperwhisper/transcript_alignment.cpp
        src/models/crisperwhisper/whisper_runtime.cpp
        src/models/crisperwhisper/session.cpp
    INCLUDES
        engine/models/crisperwhisper/model.h
    LOADERS
        engine::models::crisperwhisper::make_crisperwhisper_loader
)

audiocpp_add_model(owsm
    SOURCES
        src/models/owsm/assets.cpp
        src/models/owsm/runtime.cpp
        src/models/owsm/session.cpp
    INCLUDES
        engine/models/owsm/model.h
    LOADERS
        engine::models::owsm::make_owsm_loader
)

audiocpp_add_model(owsm_ctc
    SOURCES
        src/models/owsm_ctc/assets.cpp
        src/models/owsm_ctc/runtime.cpp
        src/models/owsm_ctc/session.cpp
    INCLUDES
        engine/models/owsm_ctc/model.h
    LOADERS
        engine::models::owsm_ctc::make_owsm_ctc_loader
)

audiocpp_add_model(cohere_asr
    SOURCES
        src/models/cohere_asr/assets.cpp
        src/models/cohere_asr/weights.cpp
        src/models/cohere_asr/runtime.cpp
        src/models/cohere_asr/session.cpp
    INCLUDES
        engine/models/cohere_asr/model.h
    LOADERS
        engine::models::cohere_asr::make_cohere_asr_loader
)

audiocpp_add_model(canary_asr
    SOURCES
        src/models/canary_asr/assets.cpp
        src/models/canary_asr/weights.cpp
        src/models/canary_asr/runtime.cpp
        src/models/canary_asr/session.cpp
    INCLUDES
        engine/models/canary_asr/model.h
    LOADERS
        engine::models::canary_asr::make_canary_asr_loader
)

audiocpp_add_model(moonshine_asr
    SOURCES
        src/models/moonshine_asr/assets.cpp
        src/models/moonshine_asr/runtime.cpp
        src/models/moonshine_asr/session.cpp
        src/models/moonshine_asr/weights.cpp
    INCLUDES
        engine/models/moonshine_asr/session.h
        engine/models/moonshine_asr/runtime.h
    LOADERS
        engine::models::moonshine_asr::make_moonshine_asr_loader
)
if (ENGINE_ENABLE_NEON)
    target_sources(engine_model_moonshine_asr PRIVATE
        src/models/moonshine_asr/neon_decoder.cpp
        src/models/moonshine_asr/neon_weights.cpp)
endif()

audiocpp_add_model(qwen3_tts
    SOURCES
        src/models/qwen3_tts/assets.cpp
        src/models/qwen3_tts/tokenizer_text.cpp
        src/models/qwen3_tts/tokenizer_speech_encoder.cpp
        src/models/qwen3_tts/tokenizer_speech_decoder.cpp
        src/models/qwen3_tts/rknn_speech_decoder.cpp
        src/models/qwen3_tts/speaker_encoder.cpp
        src/models/qwen3_tts/talker.cpp
        src/models/qwen3_tts/prompt_tts_voice_clone.cpp
        src/models/qwen3_tts/prompt_tts_voice_design.cpp
        src/models/qwen3_tts/prompt_tts_custom_voice.cpp
        src/models/qwen3_tts/session.cpp
        src/models/qwen3_tts/loader.cpp
    INCLUDES
        engine/models/qwen3_tts/loader.h
    LOADERS
        engine::models::qwen3_tts::make_qwen3_tts_loader
)

audiocpp_add_model(vieneu_v3_turbo
    SOURCES
        src/community_models/vieneu_v3_turbo/assets.cpp
        src/community_models/vieneu_v3_turbo/tokenizer_text.cpp
        src/community_models/vieneu_v3_turbo/tokenizer_speech_encoder.cpp
        src/community_models/vieneu_v3_turbo/tokenizer_speech_decoder.cpp
        src/community_models/vieneu_v3_turbo/speaker_encoder.cpp
        src/community_models/vieneu_v3_turbo/talker.cpp
        src/community_models/vieneu_v3_turbo/prompt_tts_voice_clone.cpp
        src/community_models/vieneu_v3_turbo/session.cpp
        src/community_models/vieneu_v3_turbo/loader.cpp
        src/community_models/vieneu_v3_turbo/frame_cap.cpp
        src/community_models/vieneu_v3_turbo/orchestration.cpp
        src/community_models/vieneu_v3_turbo/text_frontend.cpp
    INCLUDES
        engine/community_models/vieneu_v3_turbo/loader.h
    LOADERS
        engine::models::vieneu_v3_turbo::make_vieneu_v3_turbo_loader
)
if(MSVC)
    target_compile_options(engine_model_vieneu_v3_turbo PRIVATE /utf-8)
endif()

audiocpp_add_model(qwen3_asr
    SOURCES
        src/models/qwen3_asr/assets.cpp
        src/models/qwen3_asr/tokenizer_text.cpp
        src/models/qwen3_asr/frontend_whisper.cpp
        src/models/qwen3_asr/thinker.cpp
        src/models/qwen3_asr/prompt_asr.cpp
        src/models/qwen3_asr/postprocess.cpp
        src/models/qwen3_asr/session.cpp
        src/models/qwen3_asr/loader.cpp
    INCLUDES
        engine/models/qwen3_asr/loader.h
    LOADERS
        engine::models::qwen3_asr::make_qwen3_asr_loader
    DEPENDS
        qwen3_forced_aligner
)

audiocpp_add_model(confucius4_r2t2
    SOURCES
        src/community_models/confucius4_r2t2/assets.cpp
        src/community_models/confucius4_r2t2/text_postprocess.cpp
        src/community_models/confucius4_r2t2/tokenizer_text.cpp
        src/community_models/confucius4_r2t2/frontend_whisper.cpp
        src/community_models/confucius4_r2t2/audio_encoder.cpp
        src/community_models/confucius4_r2t2/thinker.cpp
        src/community_models/confucius4_r2t2/prompt_asr.cpp
        src/community_models/confucius4_r2t2/session.cpp
    INCLUDES
        engine/community_models/confucius4_r2t2/session.h
    LOADERS
        engine::community_models::confucius4_r2t2::make_confucius4_r2t2_loader
)

audiocpp_add_model(qwen3_forced_aligner
    SOURCES
        src/models/qwen3_forced_aligner/processor.cpp
        src/models/qwen3_forced_aligner/session.cpp
        src/models/qwen3_forced_aligner/loader.cpp
    INCLUDES
        engine/models/qwen3_forced_aligner/loader.h
    LOADERS
        engine::models::qwen3_forced_aligner::make_qwen3_forced_aligner_loader
    DEPENDS
        qwen3_asr
)

audiocpp_add_model(mms_forced_aligner
    SOURCES
        src/community_models/mms_forced_aligner/assets.cpp
        src/community_models/mms_forced_aligner/ctc_alignment.cpp
        src/community_models/mms_forced_aligner/emissions.cpp
        src/community_models/mms_forced_aligner/session.cpp
        src/community_models/mms_forced_aligner/text_processor.cpp
    INCLUDES
        engine/community_models/mms_forced_aligner/session.h
    LOADERS
        engine::community_models::mms_forced_aligner::make_mms_forced_aligner_loader
)

audiocpp_add_model(sortformer_diar
    SOURCES
        src/models/sortformer_diar/assets.cpp
        src/models/sortformer_diar/frontend.cpp
        src/models/sortformer_diar/graph.cpp
        src/models/sortformer_diar/modules.cpp
        src/models/sortformer_diar/postprocess.cpp
        src/models/sortformer_diar/session.cpp
    INCLUDES
        engine/models/sortformer_diar/session.h
    LOADERS
        engine::models::sortformer_diar::make_sortformer_diar_loader
)

audiocpp_add_model(sortformer_diar_v2
    SOURCES
        src/community_models/sortformer_diar/assets.cpp
        src/community_models/sortformer_diar/frontend.cpp
        src/community_models/sortformer_diar/weights.cpp
        src/community_models/sortformer_diar/aosc_state.cpp
        src/community_models/sortformer_diar/stream_schedule.cpp
        src/community_models/sortformer_diar/encoder.cpp
        src/community_models/sortformer_diar/session.cpp
    INCLUDES
        engine/community_models/sortformer_diar/aosc_state.h
        engine/community_models/sortformer_diar/stream_schedule.h
        engine/community_models/sortformer_diar/encoder.h
        engine/community_models/sortformer_diar/session.h
        engine/community_models/sortformer_diar/frontend.h
        engine/community_models/sortformer_diar/weights.h
        engine/community_models/sortformer_diar/assets.h
    LOADERS
        engine::community_models::sortformer_diar::make_sortformer_diar_v2_loader
)

audiocpp_add_model(nemotron_3_diar
    SOURCES
        src/models/nemotron_3_diar/assets.cpp
        src/models/nemotron_3_diar/frontend.cpp
        src/models/nemotron_3_diar/streaming.cpp
        src/models/nemotron_3_diar/encoder.cpp
        src/models/nemotron_3_diar/session.cpp
    INCLUDES
        engine/models/nemotron_3_diar/assets.h
        engine/models/nemotron_3_diar/frontend.h
        engine/models/nemotron_3_diar/streaming.h
        engine/models/nemotron_3_diar/encoder.h
        engine/models/nemotron_3_diar/session.h
    LOADERS
        engine::models::nemotron_3_diar::make_nemotron_3_diar_loader
)

audiocpp_add_model(stable_audio
    SOURCES
        src/models/stable_audio/assets.cpp
        src/models/stable_audio/conditioner.cpp
        src/models/stable_audio/conditioner_runtime.cpp
        src/models/stable_audio/foundation/conditioner.cpp
        src/models/stable_audio/foundation/oobleck_autoencoder.cpp
        src/models/stable_audio/foundation/rf_dit.cpp
        src/models/stable_audio/foundation/session.cpp
        src/models/stable_audio/loader.cpp
        src/models/stable_audio/request.cpp
        src/models/stable_audio/rf_dit.cpp
        src/models/stable_audio/same_autoencoder.cpp
        src/models/stable_audio/sampler.cpp
        src/models/stable_audio/session.cpp
        src/models/stable_audio/t5gemma.cpp
    INCLUDES
        engine/models/stable_audio/loader.h
    LOADERS
        engine::models::stable_audio::make_stable_audio_loader
)

audiocpp_add_model(yue2
    SOURCES
        src/models/yue2/ar_runtime.cpp
        src/models/yue2/assets.cpp
        src/models/yue2/lora.cpp
        src/models/yue2/nar_runtime.cpp
        src/models/yue2/pipeline.cpp
        src/models/yue2/request.cpp
        src/models/yue2/session.cpp
        src/models/yue2/tokenizer_text.cpp
        src/models/yue2/types.cpp
    INCLUDES
        engine/models/yue2/session.h
    LOADERS
        engine::models::yue2::make_yue2_loader
)

audiocpp_add_model(sheetsage2
    SOURCES
        src/models/sheetsage/audio_frontend.cpp
        src/models/sheetsage/processing.cpp
        src/models/sheetsage/runtime.cpp
        src/models/sheetsage/session.cpp
    INCLUDES
        engine/models/sheetsage/audio_frontend.h
        engine/models/sheetsage/processing.h
        engine/models/sheetsage/types.h
        engine/models/sheetsage/runtime.h
        engine/models/sheetsage/session.h
    LOADERS
        engine::models::sheetsage::make_sheetsage2_loader
)

audiocpp_add_model(auk
    SOURCES
        src/community_models/auk/audio_conditioning.cpp
        src/community_models/auk/conditioning.cpp
        src/community_models/auk/flow.cpp
        src/community_models/auk/vae.cpp
        src/community_models/auk/session.cpp
    INCLUDES
        engine/community_models/auk/session.h
    LOADERS
        engine::models::auk::make_auk_loader
)
if(TARGET engine_model_auk)
    target_link_libraries(engine_model_auk PRIVATE yaml_vendor)
endif()

audiocpp_add_model(supertonic
    SOURCES
        src/models/supertonic/assets.cpp
        src/models/supertonic/loader.cpp
        src/models/supertonic/runtime.cpp
        src/models/supertonic/session.cpp
        src/models/supertonic/tokenizer_text.cpp
    INCLUDES
        engine/models/supertonic/loader.h
    LOADERS
        engine::models::supertonic::make_supertonic_loader
)

audiocpp_add_model(citrinet_asr
    SOURCES
        src/models/citrinet_asr/assets.cpp
        src/models/citrinet_asr/runtime.cpp
        src/models/citrinet_asr/session.cpp
    INCLUDES
        engine/models/citrinet_asr/session.h
    LOADERS
        engine::models::citrinet_asr::make_citrinet_asr_loader
)

audiocpp_add_model(granite5asr
    SOURCES
        src/community_models/granite5asr/assets.cpp
        src/community_models/granite5asr/frontend.cpp
        src/community_models/granite5asr/encoder.cpp
        src/community_models/granite5asr/session.cpp
    INCLUDES
        engine/community_models/granite5asr/session.h
    LOADERS
        engine::community_models::granite5asr::make_granite5asr_loader
    ALIASES
        granite_speech5_asr
        granite_speech
        granite_speech5_ctc
)

audiocpp_add_model(niagara_asr
    SOURCES
        src/models/niagara_asr/assets.cpp
        src/models/niagara_asr/encoder.cpp
        src/models/niagara_asr/frontend.cpp
        src/models/niagara_asr/runtime.cpp
        src/models/niagara_asr/session.cpp
        src/models/niagara_asr/weights.cpp
    INCLUDES
        engine/models/niagara_asr/assets.h
        engine/models/niagara_asr/encoder.h
        engine/models/niagara_asr/frontend.h
        engine/models/niagara_asr/runtime.h
        engine/models/niagara_asr/session.h
        engine/models/niagara_asr/weights.h
    LOADERS
        engine::models::niagara_asr::make_niagara_asr_loader
)

audiocpp_add_model(whistle_asr
    SOURCES
        src/community_models/whistle_asr/assets.cpp
        src/community_models/whistle_asr/decoder.cpp
        src/community_models/whistle_asr/encoder.cpp
        src/community_models/whistle_asr/frontend.cpp
        src/community_models/whistle_asr/graph.cpp
        src/community_models/whistle_asr/runtime.cpp
        src/community_models/whistle_asr/session.cpp
    INCLUDES
        engine/community_models/whistle_asr/session.h
    LOADERS
        engine::community_models::whistle_asr::make_whistle_asr_loader
)

audiocpp_add_model(audio8_asr
    SOURCES
        src/community_models/audio8_asr/assets.cpp
        src/community_models/audio8_asr/projector.cpp
        src/community_models/audio8_asr/thinker.cpp
        src/community_models/audio8_asr/session.cpp
    INCLUDES
        engine/community_models/audio8_asr/session.h
    LOADERS
        engine::community_models::audio8_asr::make_audio8_asr_loader
    DEPENDS
        qwen3_asr
)

audiocpp_add_model(vevo2
    SOURCES
        src/models/vevo2/ar.cpp
        src/models/vevo2/assets.cpp
        src/models/vevo2/components.cpp
        src/models/vevo2/fm.cpp
        src/models/vevo2/prompt_builder.cpp
        src/models/vevo2/session.cpp
        src/models/vevo2/tokenizer_ar.cpp
        src/models/vevo2/vocoder.cpp
        src/models/vevo2/loader.cpp
    INCLUDES
        engine/models/vevo2/loader.h
    LOADERS
        engine::models::vevo2::make_vevo2_loader
)

audiocpp_add_model(seed_vc
    SOURCES
        src/models/seed_vc/assets.cpp
        src/models/seed_vc/audio_features.cpp
        src/models/seed_vc/length_regulator.cpp
        src/models/seed_vc/astral_quantizer.cpp
        src/models/seed_vc/content_features.cpp
        src/models/seed_vc/rmvpe.cpp
        src/models/seed_vc/whisper_content.cpp
        src/models/seed_vc/v1_cfm.cpp
        src/models/seed_vc/v2_cfm.cpp
        src/models/seed_vc/session.cpp
    INCLUDES
        engine/models/seed_vc/session.h
    LOADERS
        engine::models::seed_vc::make_seed_vc_loader
)

audiocpp_add_model(meanvc2
    SOURCES
        src/models/meanvc2/assets.cpp
        src/models/meanvc2/asr_encoder.cpp
        src/models/meanvc2/audio_pipeline.cpp
        src/models/meanvc2/session.cpp
        src/models/meanvc2/speaker_encoder.cpp
        src/models/meanvc2/flow_sampler_runtime.cpp
        src/models/meanvc2/vocoder.cpp
    INCLUDES
        engine/models/meanvc2/assets.h
        engine/models/meanvc2/asr_encoder.h
        engine/models/meanvc2/audio_pipeline.h
        engine/models/meanvc2/session.h
        engine/models/meanvc2/speaker_encoder.h
        engine/models/meanvc2/flow_sampler_runtime.h
        engine/models/meanvc2/vocoder.h
    LOADERS
        engine::models::meanvc2::make_meanvc2_loader
)

audiocpp_add_model(rvc
    SOURCES
        src/models/rvc/assets.cpp
        src/models/rvc/hubert.cpp
        src/models/rvc/native_pipeline.cpp
        src/models/rvc/rmvpe.cpp
        src/models/rvc/retrieval_index.cpp
        src/models/rvc/session.cpp
        src/models/rvc/synthesizer.cpp
        src/models/rvc/voice_model.cpp
    INCLUDES
        engine/models/rvc/session.h
    LOADERS
        engine::models::rvc::make_rvc_loader
)

audiocpp_add_model(chatterbox
    SOURCES
        src/models/chatterbox/assets.cpp
        src/models/chatterbox/campplus_encoder_impl.cpp
        src/models/chatterbox/audio_features.cpp
        src/models/chatterbox/audio_processing.cpp
        src/models/chatterbox/component_weights.cpp
        src/models/chatterbox/conditionals.cpp
        src/models/chatterbox/loader.cpp
        src/models/chatterbox/s3_tokenizer_impl.cpp
        src/models/chatterbox/session.cpp
        src/models/chatterbox/t3_component.cpp
        src/models/chatterbox/t3_weights.cpp
        src/models/chatterbox/text_tokenizer.cpp
        src/models/chatterbox/tts.cpp
        src/models/chatterbox/vc.cpp
        src/models/chatterbox/voice_encoder_impl.cpp
    INCLUDES
        engine/models/chatterbox/loader.h
    LOADERS
        engine::models::chatterbox::make_chatterbox_loader
)

audiocpp_add_model(chatterbox_turbo
    SOURCES
        src/community_models/chatterbox_turbo/t3_turbo_weights.cpp
        src/community_models/chatterbox_turbo/t3_turbo_component.cpp
        src/community_models/chatterbox_turbo/s3gen_turbo.cpp
        src/community_models/chatterbox_turbo/text_tokenizer_turbo.cpp
        src/community_models/chatterbox_turbo/assets.cpp
        src/community_models/chatterbox_turbo/loader.cpp
        src/community_models/chatterbox_turbo/session.cpp
        src/community_models/chatterbox_turbo/tts.cpp
    INCLUDES
        engine/community_models/chatterbox_turbo/loader.h
    LOADERS
        engine::community_models::chatterbox_turbo::make_chatterbox_turbo_loader
)

audiocpp_add_model(ace_step
    SOURCES
        src/models/ace_step/assets.cpp
        src/models/ace_step/condition_encoder.cpp
        src/models/ace_step/cover_tokenizer.cpp
        src/models/ace_step/detokenizer.cpp
        src/models/ace_step/diffusion.cpp
        src/models/ace_step/dit_weights_runtime.cpp
        src/models/ace_step/loader.cpp
        src/models/ace_step/planner.cpp
        src/models/ace_step/pre_dit.cpp
        src/models/ace_step/prompt_builder.cpp
        src/models/ace_step/repaint.cpp
        src/models/ace_step/request_parser.cpp
        src/models/ace_step/session.cpp
        src/models/ace_step/task_route.cpp
        src/models/ace_step/text_encoder.cpp
        src/models/ace_step/tokenizer_text.cpp
        src/models/ace_step/vae_decoder.cpp
        src/models/ace_step/vae_encoder.cpp
    INCLUDES
        engine/models/ace_step/loader.h
    LOADERS
        engine::models::ace_step::make_ace_step_loader
)
audiocpp_add_model(sopro_tts
    SOURCES
        src/community_models/sopro_tts/acoustic.cpp
        src/community_models/sopro_tts/assets.cpp
        src/community_models/sopro_tts/reference.cpp
        src/community_models/sopro_tts/semantic_encoder.cpp
        src/community_models/sopro_tts/semantic_lm.cpp
        src/community_models/sopro_tts/session.cpp
        src/community_models/sopro_tts/speaker_encoder.cpp
        src/community_models/sopro_tts/streaming.cpp
        src/community_models/sopro_tts/text_tokenizer.cpp
        src/community_models/sopro_tts/vocoder.cpp
    INCLUDES
        engine/community_models/sopro_tts/session.h
    LOADERS
        engine::community_models::sopro_tts::make_sopro_tts_loader
)

audiocpp_add_model(soprano_tts
    SOURCES
        src/community_models/soprano_tts/assets.cpp
        src/community_models/soprano_tts/generator.cpp
        src/community_models/soprano_tts/session.cpp
        src/community_models/soprano_tts/text_normalizer.cpp
        src/community_models/soprano_tts/tokenizer_text.cpp
        src/community_models/soprano_tts/vocoder.cpp
    INCLUDES
        engine/community_models/soprano_tts/session.h
    LOADERS
        engine::community_models::soprano_tts::make_soprano_tts_loader
)

audiocpp_add_model(mira_tts
    SOURCES
        src/community_models/mira_tts/assets.cpp
        src/community_models/mira_tts/decoder.cpp
        src/community_models/mira_tts/generator.cpp
        src/community_models/mira_tts/processor.cpp
        src/community_models/mira_tts/prompt.cpp
        src/community_models/mira_tts/session.cpp
        src/community_models/mira_tts/speaker_encoder.cpp
    INCLUDES
        engine/community_models/mira_tts/session.h
    LOADERS
        engine::community_models::mira_tts::make_mira_tts_loader
)


audiocpp_add_model(midashenglm_gen
    SOURCES
        src/models/midashenglm_gen/assets.cpp
        src/models/midashenglm_gen/ar_runtime.cpp
        src/models/midashenglm_gen/audio_tokenizer.cpp
        src/models/midashenglm_gen/flow_runtime.cpp
        src/models/midashenglm_gen/prompt_encoder.cpp
        src/models/midashenglm_gen/session.cpp
        src/models/midashenglm_gen/tokenizer_text.cpp
    INCLUDES
        engine/models/midashenglm_gen/assets.h
        engine/models/midashenglm_gen/ar_runtime.h
        engine/models/midashenglm_gen/audio_tokenizer.h
        engine/models/midashenglm_gen/flow_runtime.h
        engine/models/midashenglm_gen/prompt_encoder.h
        engine/models/midashenglm_gen/session.h
        engine/models/midashenglm_gen/tokenizer_text.h
    LOADERS
        engine::models::midashenglm_gen::make_midashenglm_gen_loader
)

audiocpp_add_model(audiosr
    SOURCES
        src/models/audiosr/assets.cpp
        src/models/audiosr/autoencoder.cpp
        src/models/audiosr/ddim.cpp
        src/models/audiosr/frontend.cpp
        src/models/audiosr/hifigan.cpp
        src/models/audiosr/pipeline.cpp
        src/models/audiosr/session.cpp
        src/models/audiosr/unet.cpp
    INCLUDES
        engine/models/audiosr/assets.h
        engine/models/audiosr/autoencoder.h
        engine/models/audiosr/ddim.h
        engine/models/audiosr/frontend.h
        engine/models/audiosr/hifigan.h
        engine/models/audiosr/pipeline.h
        engine/models/audiosr/session.h
        engine/models/audiosr/unet.h
    LOADERS
        engine::models::audiosr::make_audiosr_loader
)

audiocpp_add_model(builtin_audio_utils
    SOURCES
        src/models/builtin_audio_utils/session.cpp
    INCLUDES
        engine/models/builtin_audio_utils/session.h
    LOADERS
        engine::models::builtin_audio_utils::make_builtin_audio_utils_loader
)

audiocpp_add_model(controlfoley
    SOURCES
        src/models/controlfoley/assets.cpp
        src/models/controlfoley/conditioning.cpp
        src/models/controlfoley/pipeline.cpp
        src/models/controlfoley/session.cpp
        src/models/controlfoley/video_processing.cpp
    INCLUDES
        engine/models/controlfoley/assets.h
        engine/models/controlfoley/conditioning.h
        engine/models/controlfoley/pipeline.h
        engine/models/controlfoley/session.h
        engine/models/controlfoley/video_processing.h
    LOADERS
        engine::models::controlfoley::make_controlfoley_loader
)

audiocpp_add_model(fireredtts3
    SOURCES
        src/models/fireredtts3/assets.cpp
        src/models/fireredtts3/ar.cpp
        src/models/fireredtts3/flow.cpp
        src/models/fireredtts3/pipeline.cpp
        src/models/fireredtts3/redae.cpp
        src/models/fireredtts3/session.cpp
        src/models/fireredtts3/tokenizer_text.cpp
    INCLUDES
        engine/models/fireredtts3/assets.h
        engine/models/fireredtts3/ar.h
        engine/models/fireredtts3/flow.h
        engine/models/fireredtts3/pipeline.h
        engine/models/fireredtts3/redae.h
        engine/models/fireredtts3/session.h
        engine/models/fireredtts3/tokenizer_text.h
    LOADERS
        engine::models::fireredtts3::make_fireredtts3_loader
)

audiocpp_add_model(firered_audio
    SOURCES
        src/models/firered_audio/assets.cpp
        src/models/firered_audio/audio_encoder.cpp
        src/models/firered_audio/flow.cpp
        src/models/firered_audio/patch_encoder.cpp
        src/models/firered_audio/pipeline.cpp
        src/models/firered_audio/redae.cpp
        src/models/firered_audio/session.cpp
        src/models/firered_audio/tokenizer.cpp
        src/models/firered_audio/understanding.cpp
    INCLUDES
        engine/models/firered_audio/assets.h
        engine/models/firered_audio/audio_encoder.h
        engine/models/firered_audio/flow.h
        engine/models/firered_audio/patch_encoder.h
        engine/models/firered_audio/pipeline.h
        engine/models/firered_audio/redae.h
        engine/models/firered_audio/session.h
        engine/models/firered_audio/tokenizer.h
        engine/models/firered_audio/understanding.h
    LOADERS
        engine::models::firered_audio::make_firered_audio_loader
)

audiocpp_add_model(index_echo
    SOURCES
        src/models/index_echo/assets.cpp
        src/models/index_echo/connector.cpp
        src/models/index_echo/mapper.cpp
        src/models/index_echo/session.cpp
        src/models/index_echo/text_translation.cpp
    INCLUDES
        engine/models/index_echo/assets.h
        engine/models/index_echo/connector.h
        engine/models/index_echo/mapper.h
        engine/models/index_echo/session.h
        engine/models/index_echo/text_translation.h
    LOADERS
        engine::models::index_echo::make_index_echo_loader
    DEPENDS
        qwen3_asr
)

audiocpp_add_model(cosyvoice3
    SOURCES
        src/models/cosyvoice3/ar.cpp
        src/models/cosyvoice3/assets.cpp
        src/models/cosyvoice3/flow.cpp
        src/models/cosyvoice3/frontend.cpp
        src/models/cosyvoice3/hift.cpp
        src/models/cosyvoice3/rknn_flow.cpp
        src/models/cosyvoice3/session.cpp
        src/models/cosyvoice3/tokenizer_text.cpp
    INCLUDES
        engine/models/cosyvoice3/ar.h
        engine/models/cosyvoice3/assets.h
        engine/models/cosyvoice3/flow.h
        engine/models/cosyvoice3/frontend.h
        engine/models/cosyvoice3/hift.h
        engine/models/cosyvoice3/session.h
        engine/models/cosyvoice3/tokenizer_text.h
    LOADERS
        engine::models::cosyvoice3::make_cosyvoice3_loader
)

audiocpp_add_model(breeze_tts
    SOURCES
        src/models/breeze_tts/assets.cpp
        src/models/breeze_tts/generator.cpp
        src/models/breeze_tts/session.cpp
        src/models/breeze_tts/speech_decoder.cpp
        src/models/breeze_tts/speech_encoder.cpp
        src/models/breeze_tts/text_encoder.cpp
        src/models/breeze_tts/tokenizer_text.cpp
    INCLUDES
        engine/models/breeze_tts/assets.h
        engine/models/breeze_tts/generator.h
        engine/models/breeze_tts/session.h
        engine/models/breeze_tts/speech_decoder.h
        engine/models/breeze_tts/speech_encoder.h
        engine/models/breeze_tts/text_encoder.h
        engine/models/breeze_tts/tokenizer_text.h
    LOADERS
        engine::models::breeze_tts::make_breeze_tts_loader
)

audiocpp_add_model(vibeasr
    SOURCES
        src/community_models/vibeasr/assets.cpp
        src/community_models/vibeasr/vae_encoder.cpp
        src/community_models/vibeasr/lm_decoder.cpp
        src/community_models/vibeasr/session.cpp
    INCLUDES
        engine/community_models/vibeasr/session.h
    LOADERS
        engine::community_models::vibeasr::make_vibeasr_loader
)

audiocpp_add_model(lfm2_audio
    SOURCES
        src/community_models/lfm2_audio/asr_inputs.cpp
        src/community_models/lfm2_audio/assets.cpp
        src/community_models/lfm2_audio/chat.cpp
        src/community_models/lfm2_audio/tokenizer.cpp
        src/community_models/lfm2_audio/audio_encoder.cpp
        src/community_models/lfm2_audio/backbone.cpp
        src/community_models/lfm2_audio/lfm2_blocks.cpp
        src/community_models/lfm2_audio/depthformer.cpp
        src/community_models/lfm2_audio/detokenizer.cpp
        src/community_models/lfm2_audio/tts.cpp
        src/community_models/lfm2_audio/interleaved.cpp
        src/community_models/lfm2_audio/session.cpp
        src/community_models/lfm2_audio/weight_stores.cpp
    INCLUDES
        engine/community_models/lfm2_audio/session.h
    LOADERS
        engine::community_models::lfm2_audio::make_lfm2_audio_loader
)

audiocpp_add_model(xtts_v2
    SOURCES
        src/community_models/xtts_v2/assets.cpp
        src/community_models/xtts_v2/frontend.cpp
        src/community_models/xtts_v2/gpt.cpp
        src/community_models/xtts_v2/loader.cpp
        src/community_models/xtts_v2/session.cpp
        src/community_models/xtts_v2/tokenizer_text.cpp
        src/community_models/xtts_v2/vocoder.cpp
    INCLUDES
        engine/community_models/xtts_v2/loader.h
    LOADERS
        engine::models::xtts_v2::make_xtts_v2_loader
)

set(AUDIOCPP_ENABLED_MODELS "")
if (AUDIOCPP_MODEL_SET STREQUAL "full")
    set(AUDIOCPP_ENABLED_MODELS ${AUDIOCPP_MODEL_TARGETS})
elseif (AUDIOCPP_MODEL_SET STREQUAL "core")
    set(AUDIOCPP_ENABLED_MODELS "")
elseif (AUDIOCPP_MODEL_SET STREQUAL "custom")
    string(REPLACE "," ";" AUDIOCPP_REQUESTED_MODELS "${AUDIOCPP_MODELS}")
    foreach(requested_model IN LISTS AUDIOCPP_REQUESTED_MODELS)
        string(STRIP "${requested_model}" requested_model)
        if (requested_model STREQUAL "")
            continue()
        endif()
        if (DEFINED AUDIOCPP_MODEL_ALIAS_${requested_model})
            list(APPEND AUDIOCPP_ENABLED_MODELS ${AUDIOCPP_MODEL_ALIAS_${requested_model}})
        else()
            message(FATAL_ERROR "Unknown AUDIOCPP_MODELS entry: ${requested_model}")
        endif()
    endforeach()
    list(REMOVE_DUPLICATES AUDIOCPP_ENABLED_MODELS)
else()
    message(FATAL_ERROR "AUDIOCPP_MODEL_SET must be full, core, or custom")
endif()

set(AUDIOCPP_REGISTRY_INCLUDE_CONTENT "")
set(AUDIOCPP_REGISTRY_LOADER_CONTENT "")
foreach(model_name IN LISTS AUDIOCPP_ENABLED_MODELS)
    foreach(include_name IN LISTS AUDIOCPP_MODEL_${model_name}_INCLUDES)
        string(APPEND AUDIOCPP_REGISTRY_INCLUDE_CONTENT "#include \"${include_name}\"\n")
    endforeach()
    foreach(loader_name IN LISTS AUDIOCPP_MODEL_${model_name}_LOADERS)
        string(APPEND AUDIOCPP_REGISTRY_LOADER_CONTENT "        ${loader_name}(),\n")
    endforeach()
endforeach()

set(AUDIOCPP_LINKED_MODELS ${AUDIOCPP_ENABLED_MODELS})
set(AUDIOCPP_DEPENDENCY_SCAN_INDEX 0)
list(LENGTH AUDIOCPP_LINKED_MODELS AUDIOCPP_LINKED_MODEL_COUNT)
while(AUDIOCPP_DEPENDENCY_SCAN_INDEX LESS AUDIOCPP_LINKED_MODEL_COUNT)
    list(GET AUDIOCPP_LINKED_MODELS ${AUDIOCPP_DEPENDENCY_SCAN_INDEX} model_name)
    foreach(dependency IN LISTS AUDIOCPP_MODEL_${model_name}_DEPENDS)
        if (NOT dependency IN_LIST AUDIOCPP_MODEL_TARGETS)
            message(FATAL_ERROR "Unknown composite dependency for ${model_name}: ${dependency}")
        endif()
        if (NOT dependency IN_LIST AUDIOCPP_LINKED_MODELS)
            list(APPEND AUDIOCPP_LINKED_MODELS ${dependency})
        endif()
    endforeach()
    math(EXPR AUDIOCPP_DEPENDENCY_SCAN_INDEX "${AUDIOCPP_DEPENDENCY_SCAN_INDEX} + 1")
    list(LENGTH AUDIOCPP_LINKED_MODELS AUDIOCPP_LINKED_MODEL_COUNT)
endwhile()
file(GENERATE
    OUTPUT "${CMAKE_CURRENT_BINARY_DIR}/generated/model_registry_includes.inc"
    CONTENT "${AUDIOCPP_REGISTRY_INCLUDE_CONTENT}")
file(GENERATE
    OUTPUT "${CMAKE_CURRENT_BINARY_DIR}/generated/model_registry_loaders.inc"
    CONTENT "${AUDIOCPP_REGISTRY_LOADER_CONTENT}")

set(AUDIOCPP_RUNTIME_OBJECTS
    $<TARGET_OBJECTS:engine_core>
    $<TARGET_OBJECTS:engine_model_silero_vad>
    $<TARGET_OBJECTS:engine_model_marblenet_vad>
)
if (ENGINE_ENABLE_NEON)
    list(APPEND AUDIOCPP_RUNTIME_OBJECTS $<TARGET_OBJECTS:engine_neon>)
endif()
set(AUDIOCPP_MODEL_OBJECT_LIBRARIES "")
foreach(model_name IN LISTS AUDIOCPP_LINKED_MODELS)
    list(APPEND AUDIOCPP_RUNTIME_OBJECTS $<TARGET_OBJECTS:engine_model_${model_name}>)
    list(APPEND AUDIOCPP_MODEL_OBJECT_LIBRARIES ${AUDIOCPP_MODEL_${model_name}_OBJECT_LIBRARIES})
endforeach()
list(REMOVE_DUPLICATES AUDIOCPP_MODEL_OBJECT_LIBRARIES)
foreach(object_library IN LISTS AUDIOCPP_MODEL_OBJECT_LIBRARIES)
    list(APPEND AUDIOCPP_RUNTIME_OBJECTS $<TARGET_OBJECTS:${object_library}>)
endforeach()

message(STATUS "audio.cpp model composite: ${AUDIOCPP_MODEL_SET} selected [${AUDIOCPP_ENABLED_MODELS}] linked [${AUDIOCPP_LINKED_MODELS}]")
add_library(engine_runtime STATIC
    src/framework/runtime/registry.cpp
    ${AUDIOCPP_RUNTIME_OBJECTS}
)
if (ENGINE_ENABLE_NEON)
    target_compile_definitions(engine_runtime PUBLIC ENGINE_HAVE_NEON=1)
endif()

add_library(cjson_vendor STATIC
    external/cJSON/cJSON.c
)

# cJSON annotates every CJSON_PUBLIC function with __declspec(dllexport) on
# Windows unless told otherwise (external/cJSON/cJSON.h:59-66 defaults
# CJSON_EXPORT_SYMBOLS on). Those annotations live in the object files, so any
# DLL that absorbs this static library re-exports all 78 of them -- dllexport is
# additive, it says "also export this", never "export only this". Measured: a
# shared build exported 147 symbols where 69 were intended.
#
# No DLL is built from this tree by default, so this is latent rather than
# broken today; AUDIOCPP_BUILD_C_API is the first target to expose it.
# CJSON_HIDE_SYMBOLS is cJSON's own mechanism for exactly this case and selects
# a plain declaration with no storage class. PUBLIC rather than PRIVATE so every
# translation unit that includes cJSON.h agrees on the linkage.
target_compile_definitions(cjson_vendor PUBLIC CJSON_HIDE_SYMBOLS)

target_include_directories(cjson_vendor PUBLIC
    ${CMAKE_CURRENT_SOURCE_DIR}/external/cJSON
)

# Native package management is deliberately opt-in. Normal CLI and server
# builds do not configure an HTTP/TLS dependency. When enabled, package specs
# remain embedded so the standalone manager and managed WebUI do not require a
# source checkout or an external model_specs directory.
if (AUDIOCPP_BUILD_NATIVE_MODEL_MANAGER)
    add_subdirectory(external/cpp-httplib)
    add_library(audiocpp_package_manager STATIC
        src/framework/package_manager/manager.cpp
        src/framework/io/json.cpp
        src/framework/io/filesystem.cpp
    )
    target_include_directories(audiocpp_package_manager PUBLIC
        ${CMAKE_CURRENT_SOURCE_DIR}/include
    )
    target_include_directories(audiocpp_package_manager PRIVATE
        "${CMAKE_CURRENT_BINARY_DIR}/generated"
    )
    target_link_libraries(audiocpp_package_manager PRIVATE
        cjson_vendor
        audiocpp_cpp_httplib
    )
endif()

add_library(yaml_vendor STATIC
    external/libyaml/src/api.c
    external/libyaml/src/reader.c
    external/libyaml/src/scanner.c
    external/libyaml/src/parser.c
    external/libyaml/src/loader.c
)

target_include_directories(yaml_vendor PUBLIC
    ${CMAKE_CURRENT_SOURCE_DIR}/external/libyaml/include
)

target_compile_definitions(yaml_vendor PUBLIC ${AUDIOCPP_LIBYAML_COMPILE_DEFINITIONS})

target_include_directories(engine_runtime PUBLIC
    ${CMAKE_CURRENT_SOURCE_DIR}/include
)
target_include_directories(engine_runtime PRIVATE "${CMAKE_CURRENT_BINARY_DIR}/generated")

target_include_directories(engine_runtime PRIVATE
    ${CMAKE_CURRENT_SOURCE_DIR}/external/sentencepiece/src
    ${CMAKE_CURRENT_SOURCE_DIR}/external/llama_tokenizer
)

target_link_libraries(engine_runtime PUBLIC ggml)
find_package(Threads REQUIRED)
target_link_libraries(engine_runtime PRIVATE Threads::Threads ${CMAKE_DL_LIBS})
target_link_libraries(engine_runtime PRIVATE sentencepiece cjson_vendor yaml_vendor)
if(AUDIOCPP_STATIC_ESPEAK)
    include(tools/cmake/espeak_static.cmake)
endif()
if (AUDIOCPP_HIP_STRIX_HALO_OPTIMIZATIONS_ACTIVE)
    target_compile_definitions(engine_runtime PRIVATE ENGINE_HIP_STRIX_HALO_OPTIMIZATIONS=1)
endif()
if (ENGINE_ENABLE_RKNN)
    target_compile_definitions(engine_runtime PRIVATE ENGINE_HAVE_RKNN=1)
    if (ENGINE_RKNN_INCLUDE_DIR)
        target_include_directories(engine_runtime PUBLIC ${ENGINE_RKNN_INCLUDE_DIR})
    endif()
    target_link_libraries(engine_runtime PUBLIC ${ENGINE_RKNN_LIBRARY})
endif()
if (ENGINE_ENABLE_HIP)
    # ggml-hip publicly defines GGML_USE_CUDA for consumers, so engine code
    # needs this marker to tell a real CUDA build apart from a HIP build.
    target_compile_definitions(engine_core PRIVATE ENGINE_GGML_HIP_BACKEND=1)
endif()

# Fish Fast-AR emits a 4096-entry logits tensor for each acoustic codebook.
# On Linux HIP, a small device-side exact top-k helper keeps only the compact
# sampling candidates crossing back to the host. Built as a proper HIP target
# so CMake tracks its HIP runtime dependency; other backends stay unchanged.
if (ENGINE_ENABLE_HIP AND UNIX AND NOT APPLE)
    add_subdirectory(src/models/fish_audio)
    target_link_libraries(engine_runtime PRIVATE engine_fish_audio_hip_sampler)
    target_compile_definitions(engine_model_fish_audio PRIVATE ENGINE_HAS_HIP_FISH_FAST_SAMPLER=1)
endif()

if (ENGINE_ENABLE_OPENMP)
    target_link_libraries(engine_runtime PRIVATE OpenMP::OpenMP_CXX)
    if (MSVC)
        # MSVC's default /openmp implements only OpenMP 2.0 and rejects the
        # '#pragma omp simd' directives in longformer_attention.cpp (error C7660).
        # /openmp:experimental enables the OpenMP 4.0 SIMD support; it overrides the
        # /openmp added by OpenMP::OpenMP_CXX above (harmless D9025 override notice).
        # Guard on CXX: nvcc does not forward a bare /openmp:experimental to the host
        # compiler and instead treats it as a second input file, which is fatal.
        target_compile_options(engine_runtime PRIVATE $<$<COMPILE_LANGUAGE:CXX>:/openmp:experimental>)
    endif()
endif()

if (MSVC)
    target_compile_options(engine_core PRIVATE $<$<COMPILE_LANGUAGE:CXX>:/utf-8>)
    target_compile_options(engine_runtime PRIVATE $<$<COMPILE_LANGUAGE:CXX>:/utf-8>)
endif()

if (ENGINE_ENABLE_CUDA AND NOT ENGINE_ENABLE_HIP)
    target_sources(engine_runtime PRIVATE
        src/framework/audio/istft_cuda_runtime.cu
        src/framework/sampling/torch_random_cuda_runtime.cu
    )
    set_source_files_properties(
        src/framework/audio/istft_cuda_runtime.cu
        src/framework/sampling/torch_random_cuda_runtime.cu
        PROPERTIES
        CUDA_STANDARD 17
        CUDA_STANDARD_REQUIRED ON
    )
    target_compile_definitions(engine_core PRIVATE ENGINE_HAS_CUDA_ISTFT=1 ENGINE_HAS_CUDA_TORCH_RANDOM=1)
    target_compile_definitions(engine_runtime PUBLIC ENGINE_HAS_CUDA_ISTFT=1 ENGINE_HAS_CUDA_TORCH_RANDOM=1)
    target_link_libraries(engine_runtime PRIVATE CUDA::cudart CUDA::cufft)
endif()

# Windows decodes narrow strings - argv, file paths, console output - with the
# process code page, and the default is a legacy one: a path holding "Thiền"
# then names a file that is not there, and converting such a path back to a
# narrow string throws. This manifest asks the process for UTF-8 instead
# (Windows 10 1903 and later), which is what every string in this codebase
# already is.
function(audiocpp_use_utf8_codepage target)
    if (MSVC)
        target_sources(${target} PRIVATE "${CMAKE_CURRENT_SOURCE_DIR}/tools/cmake/audiocpp_utf8.manifest")
    endif()
endfunction()

file(GLOB AUDIOCPP_RECIPE_PLUGIN_SOURCES CONFIGURE_DEPENDS
    "${CMAKE_CURRENT_SOURCE_DIR}/app/recipe/plugins/*.cpp")
set(AUDIOCPP_RECIPE_SOURCES
    app/recipe/recipe.cpp
    ${AUDIOCPP_RECIPE_PLUGIN_SOURCES})

add_executable(audiocpp_cli
    app/common/base64.cpp
    app/common/build_info.cpp
    app/cli/main.cpp
    app/cli/args.cpp
    app/cli/batch.cpp
    app/cli/request.cpp
    app/streaming/pcm_source.cpp
    app/streaming/streaming.cpp
    ${AUDIOCPP_RECIPE_SOURCES}
    app/recipe/cli_recipe.cpp
    app/workflow/execution.cpp
    app/workflow/file_sink.cpp
    app/workflow/pipeline.cpp
    app/workflow/workflow.cpp
)

target_link_libraries(audiocpp_cli PRIVATE engine_runtime ggml)
audiocpp_use_utf8_codepage(audiocpp_cli)
if(AUDIOCPP_STATIC_ESPEAK)
    audiocpp_stage_espeak(audiocpp_cli)
endif()
target_include_directories(audiocpp_cli PRIVATE "${CMAKE_CURRENT_BINARY_DIR}/generated")
# MinGW selects the wmain() entry point only when linked with -municode;
# without it the CLI fails to link (undefined reference / ld error 5).
# MSVC detects wmain natively, so this is MinGW-only. Link option only:
# -municode also defines UNICODE at compile time, so passing it link-only
# keeps the source compiled with no macro side effects.
if (MINGW)
    target_link_options(audiocpp_cli PRIVATE -municode)
endif()
if (ENGINE_ENABLE_OPENMP)
    target_link_libraries(audiocpp_cli PRIVATE OpenMP::OpenMP_CXX)
endif()

# ---------------------------------------------------------------------------
# C ABI (opt-in)
#
# A thin facade over engine::runtime for embedding audio.cpp in-process from
# languages that can call C. Purely additive: nothing above this point changes
# when AUDIOCPP_BUILD_C_API is OFF, which is the default.
# ---------------------------------------------------------------------------
if (AUDIOCPP_BUILD_C_API)
    add_library(audiocpp SHARED src/capi/audiocpp.cpp)
    target_link_libraries(audiocpp PRIVATE engine_runtime ggml)
    target_include_directories(audiocpp PUBLIC
        $<BUILD_INTERFACE:${CMAKE_CURRENT_SOURCE_DIR}/include>
    )
    target_compile_definitions(audiocpp
        PRIVATE
            AUDIOCPP_BUILD_DLL
            AUDIOCPP_VERSION_STRING="${AUDIOCPP_VERSION}"
        # Anything linking this gets __declspec(dllimport) on Windows without
        # having to know to ask for it. No effect elsewhere.
        INTERFACE
            AUDIOCPP_USE_DLL
    )
    set_target_properties(audiocpp PROPERTIES
        ARCHIVE_OUTPUT_DIRECTORY "${CMAKE_BINARY_DIR}/bin"
        VERSION 0.1.0
        SOVERSION 0
        C_VISIBILITY_PRESET hidden
        CXX_VISIBILITY_PRESET hidden
        VISIBILITY_INLINES_HIDDEN ON
    )
    if (APPLE)
        set_target_properties(audiocpp PROPERTIES
            INSTALL_RPATH "@loader_path" BUILD_WITH_INSTALL_RPATH ON)
    elseif (UNIX)
        set_target_properties(audiocpp PROPERTIES
            INSTALL_RPATH "$ORIGIN" BUILD_WITH_INSTALL_RPATH ON)
    endif()
    # The visibility presets above only reach this target's own sources. The
    # static libraries linked in (ggml, cJSON, sentencepiece, ...) are built
    # without -fvisibility=hidden, so without an explicit export list they are
    # re-exported from libaudiocpp -- measured at 2767 dynamic symbols where the
    # header declares 69 -- and can collide with whatever the host application
    # has already loaded. Restrict the surface to audiocpp_*.
    #
    # Windows has no equivalent of these allowlists: __declspec(dllexport) is
    # additive, so a DLL's surface is whatever its own annotations plus its
    # dependencies' annotations add up to. Exclusivity there comes from making
    # sure no dependency annotates its symbols, which is why cjson_vendor is
    # built with CJSON_HIDE_SYMBOLS above.
    if (APPLE)
        target_link_options(audiocpp PRIVATE
            "-Wl,-exported_symbols_list,${CMAKE_CURRENT_SOURCE_DIR}/src/capi/audiocpp.symbols")
        set_property(TARGET audiocpp APPEND PROPERTY
            LINK_DEPENDS "${CMAKE_CURRENT_SOURCE_DIR}/src/capi/audiocpp.symbols")
    elseif (NOT WIN32)
        target_link_options(audiocpp PRIVATE
            "-Wl,--version-script=${CMAKE_CURRENT_SOURCE_DIR}/src/capi/audiocpp.map"
            "-Wl,--exclude-libs,ALL")
        set_property(TARGET audiocpp APPEND PROPERTY
            LINK_DEPENDS "${CMAKE_CURRENT_SOURCE_DIR}/src/capi/audiocpp.map")
    endif()
    if (ENGINE_ENABLE_OPENMP)
        target_link_libraries(audiocpp PRIVATE OpenMP::OpenMP_CXX)
    endif()
    if(AUDIOCPP_STATIC_ESPEAK)
        audiocpp_stage_espeak(audiocpp)
    endif()

    # enable_testing() is called again further down for the framework test
    # suites; repeating it here is harmless, and it is what lets this test
    # register without depending on ENGINE_BUILD_TESTS.
    enable_testing()

    # Compiled as C, not C++: the point of the test is that the header is
    # consumable by a C compiler with no C++ runtime on the caller's side.
    add_executable(audiocpp_c_api_path_test tests/capi/path_test.c)
    target_link_libraries(audiocpp_c_api_path_test PRIVATE audiocpp)
    set_target_properties(audiocpp_c_api_path_test PROPERTIES C_STANDARD 99)
    add_test(NAME audiocpp_c_api_path
        COMMAND audiocpp_c_api_path_test
            "${CMAKE_CURRENT_SOURCE_DIR}/assets/framework/models/silero_vad"
            "${CMAKE_CURRENT_SOURCE_DIR}/assets/resources/sample_16k.wav")
    if (GGML_BACKEND_DL)
        set(AUDIOCPP_C_API_LAYOUT_DIR "${CMAKE_CURRENT_BINARY_DIR}/capi-library-layout")
        set(AUDIOCPP_C_API_LAYOUT_ENV
            "LD_LIBRARY_PATH=$<TARGET_FILE_DIR:audiocpp>:$ENV{LD_LIBRARY_PATH}")
        if (WIN32)
            set(AUDIOCPP_C_API_LAYOUT_ENV "PATH=$<TARGET_FILE_DIR:audiocpp>;$ENV{PATH}")
        elseif (APPLE)
            set(AUDIOCPP_C_API_LAYOUT_ENV
                "DYLD_LIBRARY_PATH=$<TARGET_FILE_DIR:audiocpp>:$ENV{DYLD_LIBRARY_PATH}")
        endif()
        add_custom_command(TARGET audiocpp_c_api_path_test POST_BUILD
            COMMAND ${CMAKE_COMMAND} -E make_directory "${AUDIOCPP_C_API_LAYOUT_DIR}"
            COMMAND ${CMAKE_COMMAND} -E copy "$<TARGET_FILE:audiocpp_c_api_path_test>"
                "${AUDIOCPP_C_API_LAYOUT_DIR}/$<TARGET_FILE_NAME:audiocpp_c_api_path_test>")
        add_test(NAME audiocpp_c_api_library_layout
            COMMAND ${CMAKE_COMMAND} -E env
                "${AUDIOCPP_C_API_LAYOUT_ENV}"
                "${AUDIOCPP_C_API_LAYOUT_DIR}/$<TARGET_FILE_NAME:audiocpp_c_api_path_test>"
                "${CMAKE_CURRENT_SOURCE_DIR}/assets/framework/models/silero_vad"
                "${CMAKE_CURRENT_SOURCE_DIR}/assets/resources/sample_16k.wav")
        set_tests_properties(audiocpp_c_api_library_layout PROPERTIES
            WORKING_DIRECTORY "${CMAKE_CURRENT_SOURCE_DIR}")
    endif()

    # Real families across task types. Needs downloaded models, so it reports
    # SKIP rather than failure when AUDIOCPP_C_API_MODEL_ROOT is unset or a
    # package is absent -- a checkout with no downloads still goes green.
    add_executable(audiocpp_c_api_model_test tests/capi/model_test.c)
    target_link_libraries(audiocpp_c_api_model_test PRIVATE audiocpp)
    set_target_properties(audiocpp_c_api_model_test PROPERTIES C_STANDARD 99)
    # Source separation dominates this test's wall time and scales with threads
    # (measured 191s at 4, 113s at 16), and thread count does not change results
    # -- verified bit-identical stems at both. So use the host's cores.
    include(ProcessorCount)
    ProcessorCount(AUDIOCPP_C_API_TEST_THREADS)
    if (AUDIOCPP_C_API_TEST_THREADS EQUAL 0)
        set(AUDIOCPP_C_API_TEST_THREADS 4)
    endif()
    add_test(NAME audiocpp_c_api_model
        COMMAND audiocpp_c_api_model_test
            "${AUDIOCPP_C_API_MODEL_ROOT}"
            "${CMAKE_CURRENT_SOURCE_DIR}/assets/resources/sample_16k.wav"
            "cpu"
            "${CMAKE_CURRENT_SOURCE_DIR}/tests/ace_step/assets/complete_source_demucs_8s.wav"
            "${AUDIOCPP_C_API_TEST_THREADS}")
    set_tests_properties(audiocpp_c_api_model PROPERTIES SKIP_RETURN_CODE 77)

    find_package(Python3 COMPONENTS Interpreter QUIET)

    # The export surface is a property three platforms reach three different
    # ways, and on Windows nothing enforces it at all -- a newly vendored
    # library that annotates its own symbols would silently widen the DLL.
    # cJSON did exactly that. So it is asserted rather than left to a comment.
    if (Python3_Interpreter_FOUND)
        # dumpbin ships beside the cl.exe that built the library, so its location
        # is already known here -- no searching, no version guessing, and it is
        # guaranteed to match the toolset that produced the binary. It is not on
        # PATH outside a Developer Command Prompt, so without this the check
        # silently skips on the one platform that has no export allowlist.
        set(AUDIOCPP_EXPORT_TOOL_ARGS "")
        if (MSVC)
            get_filename_component(AUDIOCPP_MSVC_BIN "${CMAKE_CXX_COMPILER}" DIRECTORY)
            find_program(AUDIOCPP_DUMPBIN dumpbin HINTS "${AUDIOCPP_MSVC_BIN}" NO_DEFAULT_PATH)
            if (AUDIOCPP_DUMPBIN)
                list(APPEND AUDIOCPP_EXPORT_TOOL_ARGS --symbol-tool "${AUDIOCPP_DUMPBIN}")
            endif()
            # The tool ships with the compiler, so on MSVC its absence is a real
            # problem rather than an unsupported platform. Fail loudly instead of
            # reporting a green run with the invariant unchecked.
            list(APPEND AUDIOCPP_EXPORT_TOOL_ARGS --require-tool)
        endif()
        add_test(NAME audiocpp_c_api_exports
            COMMAND "${Python3_EXECUTABLE}" "${CMAKE_CURRENT_SOURCE_DIR}/tests/capi/export_surface.py"
                --build-dir "${CMAKE_BINARY_DIR}"
                --header "${CMAKE_CURRENT_SOURCE_DIR}/include/audiocpp.h"
                ${AUDIOCPP_EXPORT_TOOL_ARGS})
        set_tests_properties(audiocpp_c_api_exports PROPERTIES SKIP_RETURN_CODE 77)
    endif()

    # The C API drives the same runtime audiocpp_cli does, so for identical
    # inputs the two must produce identical output. That is the property worth
    # testing -- not that the facade is plausible, but that it agrees. Skips
    # when the CLI or the models are absent.
    if (Python3_Interpreter_FOUND)
        add_test(NAME audiocpp_c_api_parity
            COMMAND "${Python3_EXECUTABLE}" "${CMAKE_CURRENT_SOURCE_DIR}/tests/capi/parity.py"
                --build-dir "${CMAKE_BINARY_DIR}"
                --models-root "${AUDIOCPP_C_API_MODEL_ROOT}"
                --audio "${CMAKE_CURRENT_SOURCE_DIR}/assets/resources/sample_16k.wav"
                --alt-audio "${CMAKE_CURRENT_SOURCE_DIR}/tests/ace_step/assets/complete_source_demucs_8s.wav"
                --threads "${AUDIOCPP_C_API_TEST_THREADS}")
        set_tests_properties(audiocpp_c_api_parity PROPERTIES SKIP_RETURN_CODE 77)
    endif()
endif()

set(AUDIOCPP_UI_DIST "${CMAKE_CURRENT_SOURCE_DIR}/webui/native/dist/index.html")
set_property(DIRECTORY APPEND PROPERTY CMAKE_CONFIGURE_DEPENDS "${AUDIOCPP_UI_DIST}")
if (EXISTS "${AUDIOCPP_UI_DIST}")
    file(READ "${AUDIOCPP_UI_DIST}" AUDIOCPP_UI_HEX HEX)
else()
    string(HEX "<!doctype html><title>audio.cpp</title><p>Embedded WebUI assets are unavailable.</p>" AUDIOCPP_UI_HEX)
endif()
string(REGEX REPLACE "([0-9a-f][0-9a-f])" "0x\\1," AUDIOCPP_UI_BYTES "${AUDIOCPP_UI_HEX}")
string(REPLACE "," ",\n" AUDIOCPP_UI_BYTES "${AUDIOCPP_UI_BYTES}")
configure_file(
    "${CMAKE_CURRENT_SOURCE_DIR}/app/server/audiocpp_ui_asset.h.in"
    "${CMAKE_CURRENT_BINARY_DIR}/generated/audiocpp_ui_asset.h"
    @ONLY
)

set(AUDIOCPP_DEMO_VOICE_DIR "${CMAKE_CURRENT_SOURCE_DIR}/webui/native/demo_voices")
macro(audiocpp_embed_demo_voice_file INPUT_FILE OUTPUT_VARIABLE)
    set_property(DIRECTORY APPEND PROPERTY CMAKE_CONFIGURE_DEPENDS "${INPUT_FILE}")
    file(READ "${INPUT_FILE}" _audiocpp_demo_voice_hex HEX)
    string(REGEX REPLACE "([0-9a-f][0-9a-f])" "0x\\1," ${OUTPUT_VARIABLE} "${_audiocpp_demo_voice_hex}")
    string(REPLACE "," ",\n" ${OUTPUT_VARIABLE} "${${OUTPUT_VARIABLE}}")
endmacro()
audiocpp_embed_demo_voice_file(
    "${AUDIOCPP_DEMO_VOICE_DIR}/demo_1_man.wav"
    AUDIOCPP_DEMO_1_MAN_BYTES)
audiocpp_embed_demo_voice_file(
    "${AUDIOCPP_DEMO_VOICE_DIR}/demo_2_man.wav"
    AUDIOCPP_DEMO_2_MAN_BYTES)
audiocpp_embed_demo_voice_file(
    "${AUDIOCPP_DEMO_VOICE_DIR}/demo_3_woman.wav"
    AUDIOCPP_DEMO_3_WOMAN_BYTES)
audiocpp_embed_demo_voice_file(
    "${AUDIOCPP_DEMO_VOICE_DIR}/demo_4_woman.wav"
    AUDIOCPP_DEMO_4_WOMAN_BYTES)
audiocpp_embed_demo_voice_file(
    "${AUDIOCPP_DEMO_VOICE_DIR}/prompt_text"
    AUDIOCPP_DEMO_PROMPT_TEXT_BYTES)
configure_file(
    "${CMAKE_CURRENT_SOURCE_DIR}/app/server/audiocpp_demo_voice_assets.h.in"
    "${CMAKE_CURRENT_BINARY_DIR}/generated/audiocpp_demo_voice_assets.h"
    @ONLY
)

add_executable(audiocpp_server
    app/common/base64.cpp
    app/common/build_info.cpp
    app/server/main.cpp
    app/server/config.cpp
    app/server/frontend.cpp
    app/server/http.cpp
    app/server/parallel_http.cpp
    app/server/model_memory.cpp
    app/server/multipart.cpp
    app/server/server_protocol.cpp
    app/server/server_ui.cpp
    app/server/runtime.cpp
    app/server/parallel_runtime.cpp
    app/server/ui_assets.cpp
    app/cli/args.cpp
    app/cli/request.cpp
    app/streaming/pcm_source.cpp
    app/streaming/streaming.cpp
    ${AUDIOCPP_RECIPE_SOURCES}
)

target_link_libraries(audiocpp_server PRIVATE engine_runtime ggml)
audiocpp_use_utf8_codepage(audiocpp_server)
if(AUDIOCPP_STATIC_ESPEAK)
    audiocpp_stage_espeak(audiocpp_server)
endif()
audiocpp_configure_server_frontends(audiocpp_server)
if (AUDIOCPP_BUILD_NATIVE_MODEL_MANAGER)
    target_sources(audiocpp_server PRIVATE app/server/model_installer.cpp)
    target_link_libraries(audiocpp_server PRIVATE audiocpp_package_manager)
    target_compile_definitions(audiocpp_server PRIVATE AUDIOCPP_HAS_NATIVE_MODEL_MANAGER=1)
endif()
target_include_directories(audiocpp_server PRIVATE "${CMAKE_CURRENT_BINARY_DIR}/generated")
if (WIN32)
    target_link_libraries(audiocpp_server PRIVATE ws2_32)
endif()
if (ENGINE_ENABLE_OPENMP)
    target_link_libraries(audiocpp_server PRIVATE OpenMP::OpenMP_CXX)
endif()

add_executable(audiocpp_gguf
    app/gguf/main.cpp
)
target_link_libraries(audiocpp_gguf PRIVATE engine_runtime ggml)
audiocpp_use_utf8_codepage(audiocpp_gguf)
target_include_directories(audiocpp_gguf PRIVATE "${CMAKE_CURRENT_BINARY_DIR}/generated")
if (ENGINE_ENABLE_OPENMP)
    target_link_libraries(audiocpp_gguf PRIVATE OpenMP::OpenMP_CXX)
endif()

if (AUDIOCPP_BUILD_NATIVE_MODEL_MANAGER)
    add_executable(audiocpp_model_manager
        app/model_manager/main.cpp
    )
    target_link_libraries(audiocpp_model_manager PRIVATE audiocpp_package_manager)
endif()

add_executable(model_perf
    tests/perf/model_perf.cpp
    app/cli/args.cpp
    app/cli/request.cpp
    app/common/base64.cpp
)
target_link_libraries(model_perf PRIVATE engine_runtime ggml)
audiocpp_use_utf8_codepage(model_perf)
target_include_directories(model_perf PRIVATE ${CMAKE_CURRENT_SOURCE_DIR})
if (ENGINE_ENABLE_OPENMP)
    target_link_libraries(model_perf PRIVATE OpenMP::OpenMP_CXX)
endif()

if (ENGINE_BUILD_TESTS OR ENGINE_BUILD_EXTENDED_TESTS OR ENGINE_BUILD_MODEL_TESTS)
    if (firered_vad IN_LIST AUDIOCPP_LINKED_MODELS)
        add_executable(firered_vad_probe tests/firered_vad/probe.cpp)
        target_link_libraries(firered_vad_probe PRIVATE engine_runtime ggml)
        add_executable(firered_vad_rules_test tests/firered_vad/test_rules.cpp)
        target_link_libraries(firered_vad_rules_test PRIVATE engine_runtime ggml)
        add_test(NAME firered_vad_rules_test COMMAND firered_vad_rules_test)
    endif()
    if (moss_tts_nano IN_LIST AUDIOCPP_LINKED_MODELS)
        # (moved next to the other moss_tts_nano tests)
    endif()
    if ("kitten_tts2" IN_LIST AUDIOCPP_LINKED_MODELS)
        add_executable(audiocpp_kitten_tts2_prepare_voices tools/community_models/prepare_kitten_tts2_voices.cpp)
        audiocpp_use_utf8_codepage(audiocpp_kitten_tts2_prepare_voices)
        target_link_libraries(audiocpp_kitten_tts2_prepare_voices PRIVATE engine_runtime)
        add_executable(kitten_tts2_session_probe tests/kitten_tts2/session_probe.cpp)
        audiocpp_use_utf8_codepage(kitten_tts2_session_probe)
        target_link_libraries(kitten_tts2_session_probe PRIVATE engine_runtime)
        add_executable(kitten_tts2_components_probe tests/kitten_tts2/components_probe.cpp)
        audiocpp_use_utf8_codepage(kitten_tts2_components_probe)
        target_link_libraries(kitten_tts2_components_probe PRIVATE engine_runtime)
    endif()
    add_executable(yue2_vae_parity_probe
        tests/yue2/yue2_vae_parity_probe.cpp
    )
    target_link_libraries(yue2_vae_parity_probe PRIVATE engine_runtime ggml)
    if (ENGINE_ENABLE_OPENMP)
        target_link_libraries(yue2_vae_parity_probe PRIVATE OpenMP::OpenMP_CXX)
    endif()

    # The NAR probe uses the YuE2 model runtime, which only exists when the
    # model composite links it; the VAE probe above only needs framework code.
    if ("yue2" IN_LIST AUDIOCPP_LINKED_MODELS)
        add_executable(yue2_nar_parity_probe
            tests/yue2/yue2_nar_parity_probe.cpp
        )
        target_link_libraries(yue2_nar_parity_probe PRIVATE engine_runtime ggml)
        if (ENGINE_ENABLE_OPENMP)
            target_link_libraries(yue2_nar_parity_probe PRIVATE OpenMP::OpenMP_CXX)
        endif()
    endif()

    add_executable(sheetsage2_decoder_parity_probe
        src/models/sheetsage/runtime.cpp
        tests/yue2/sheetsage2_decoder_parity_probe.cpp
    )
    target_link_libraries(sheetsage2_decoder_parity_probe PRIVATE engine_runtime ggml)
    if (ENGINE_ENABLE_OPENMP)
        target_link_libraries(sheetsage2_decoder_parity_probe PRIVATE OpenMP::OpenMP_CXX)
    endif()
endif()

if (ENGINE_BUILD_WARMBENCH)
    function(add_engine_warmbench target_name source_file)
        add_executable(${target_name}
            ${source_file}
        )
        target_link_libraries(${target_name} PRIVATE engine_runtime ggml)
        if (ENGINE_ENABLE_OPENMP)
            target_link_libraries(${target_name} PRIVATE OpenMP::OpenMP_CXX)
        endif()
    endfunction()

    add_engine_warmbench(campplus_shared_default_probe tests/glm_tts/campplus_shared_default_probe.cpp)
    add_engine_warmbench(chatterbox_warm_bench tests/chatterbox/chatterbox_warm_bench.cpp)
    if(kokoro_tts IN_LIST AUDIOCPP_LINKED_MODELS)
        add_engine_warmbench(kokoro_tts_warm_bench tests/kokoro_tts/kokoro_tts_warm_bench.cpp)
    endif()
    add_engine_warmbench(citrinet_asr_warm_bench tests/citrinet_asr/citrinet_asr_warm_bench.cpp)
    add_engine_warmbench(confucius4_tts_warm_bench tests/confucius4_tts/confucius4_tts_warm_bench.cpp)
    add_engine_warmbench(controlfoley_warm_bench tests/controlfoley/controlfoley_warm_bench.cpp)
    add_engine_warmbench(fun_asr_nano_warm_bench tests/fun_asr_nano/fun_asr_nano_warm_bench.cpp)
    add_engine_warmbench(granite5asr_warm_bench tests/granite5asr/granite5asr_warm_bench.cpp)
    target_compile_definitions(granite5asr_warm_bench PRIVATE
        ENGINE_REPO_ROOT="${CMAKE_CURRENT_SOURCE_DIR}"
    )
    add_engine_warmbench(higgs_audio_stt_warm_bench tests/higgs_audio_stt/higgs_audio_stt_warm_bench.cpp)
    add_engine_warmbench(higgs_audio_tts_warm_bench tests/higgs_audio_tts/higgs_audio_tts_warm_bench.cpp)
    add_engine_warmbench(hviske_asr_warm_bench tests/hviske_asr/hviske_asr_warm_bench.cpp)
    add_engine_warmbench(index_tts2_warm_bench tests/index_tts2/index_tts2_warm_bench.cpp)
    add_engine_warmbench(irodori_tts_warm_bench tests/irodori_tts/irodori_tts_warm_bench.cpp)
    add_engine_warmbench(marblenet_vad_warm_bench tests/marblenet_vad/marblenet_vad_warm_bench.cpp)
    add_engine_warmbench(miocodec_warm_bench tests/miocodec/miocodec_warm_bench.cpp)
    add_engine_warmbench(mira_tts_warm_bench tests/mira_tts/mira_tts_warm_bench.cpp)
    add_engine_warmbench(muscriptor_warm_bench tests/muscriptor/muscriptor_warm_bench.cpp)
    add_engine_warmbench(moss_tts_nano_warm_bench tests/moss_tts_nano/moss_tts_nano_warm_bench.cpp)
    add_engine_warmbench(moss_tts_local_warm_bench tests/moss_tts_local/moss_tts_local_warm_bench.cpp)
    add_engine_warmbench(nemotron_asr_warm_bench tests/nemotron_asr/nemotron_asr_warm_bench.cpp)
    add_engine_warmbench(neutts_warm_bench tests/neutts/neutts_warm_bench.cpp)
    add_engine_warmbench(omnivoice_warm_bench tests/omnivoice/omnivoice_warm_bench.cpp)
    add_engine_warmbench(outetts_warm_bench tests/outetts/outetts_warm_bench.cpp)
    add_engine_warmbench(parakeet_warm_bench tests/parakeet_tdt/parakeet_warm_bench.cpp)
    add_engine_warmbench(glm_tts_tokenizer_probe tests/glm_tts/glm_tts_tokenizer_probe.cpp)
    add_engine_warmbench(glm_tts_assets_probe tests/glm_tts/glm_tts_assets_probe.cpp)
    add_engine_warmbench(glm_tts_prompt_probe tests/glm_tts/glm_tts_prompt_probe.cpp)
    add_engine_warmbench(glm_tts_speech_tokenizer_probe tests/glm_tts/glm_tts_speech_tokenizer_probe.cpp)
    add_engine_warmbench(glm_tts_flow_probe tests/glm_tts/glm_tts_flow_probe.cpp)
    add_engine_warmbench(glm_tts_flow_conditioned_probe tests/glm_tts/glm_tts_flow_conditioned_probe.cpp)
    add_engine_warmbench(glm_tts_frontend_probe tests/glm_tts/glm_tts_frontend_probe.cpp)
    add_engine_warmbench(glm_tts_campplus_probe tests/glm_tts/glm_tts_campplus_probe.cpp)
    add_engine_warmbench(glm_tts_llama_probe tests/glm_tts/glm_tts_llama_probe.cpp)
    add_engine_warmbench(glm_tts_warm_bench tests/glm_tts/glm_tts_warm_bench.cpp)
    add_engine_warmbench(kroko_asr_encoder_probe tests/kroko_asr/kroko_asr_encoder_probe.cpp)
    add_engine_warmbench(kroko_asr_frontend_probe tests/kroko_asr/kroko_asr_frontend_probe.cpp)
    add_engine_warmbench(kroko_asr_streaming_probe tests/kroko_asr/kroko_asr_streaming_probe.cpp)
    add_engine_warmbench(kroko_asr_zipformer_probe tests/kroko_asr/kroko_asr_zipformer_probe.cpp)
    add_engine_warmbench(pocket_tts_warm_bench tests/pocket_tts/pocket_tts_warm_bench.cpp)
    add_engine_warmbench(qwen3_asr_warm_bench tests/qwen3_asr/qwen3_asr_warm_bench.cpp)
    add_engine_warmbench(qwen3_forced_aligner_warm_bench tests/qwen3_forced_aligner/qwen3_forced_aligner_warm_bench.cpp)
    add_engine_warmbench(mms_forced_aligner_warm_bench tests/mms_forced_aligner/mms_forced_aligner_warm_bench.cpp)
    add_engine_warmbench(qwen3_tts_warm_bench tests/qwen3_tts/qwen3_tts_warm_bench.cpp)
    add_engine_warmbench(seed_vc_warm_bench tests/seed_vc/seed_vc_warm_bench.cpp)
    add_engine_warmbench(silero_vad_warm_bench tests/silero_vad/silero_vad_warm_bench.cpp)
    add_engine_warmbench(soprano_warm_bench tests/soprano_tts/soprano_warm_bench.cpp)
    target_compile_definitions(soprano_warm_bench PRIVATE
        ENGINE_REPO_ROOT="${CMAKE_CURRENT_SOURCE_DIR}"
    )
    add_engine_warmbench(sopro_probe tests/sopro_tts/sopro_probe.cpp)
    add_engine_warmbench(sortformer_diar_warm_bench tests/sortformer_diar/sortformer_diar_warm_bench.cpp)
    add_engine_warmbench(supertonic_warm_bench tests/supertonic/supertonic_warm_bench.cpp)
    add_engine_warmbench(vevo2_warm_bench tests/vevo2/vevo2_warm_bench.cpp)
    add_engine_warmbench(vibevoice_asr_warm_bench tests/vibevoice_asr/vibevoice_asr_warm_bench.cpp)
    add_engine_warmbench(voxtral_realtime_warm_bench tests/voxtral_realtime/voxtral_realtime_warm_bench.cpp)
    add_engine_warmbench(voxcpm2_warm_bench tests/voxcpm2/voxcpm2_warm_bench.cpp)
endif()

if (ENGINE_BUILD_TESTS OR ENGINE_BUILD_EXTENDED_TESTS OR ENGINE_BUILD_MODEL_TESTS)
    enable_testing()

    if(cohere_asr IN_LIST AUDIOCPP_LINKED_MODELS AND
       canary_asr IN_LIST AUDIOCPP_LINKED_MODELS AND
       citrinet_asr IN_LIST AUDIOCPP_LINKED_MODELS AND
       parakeet_tdt IN_LIST AUDIOCPP_LINKED_MODELS AND
       sortformer_diar IN_LIST AUDIOCPP_LINKED_MODELS AND
       sortformer_diar_v2 IN_LIST AUDIOCPP_LINKED_MODELS AND
       hviske_asr IN_LIST AUDIOCPP_LINKED_MODELS AND
       nemotron_asr IN_LIST AUDIOCPP_LINKED_MODELS AND
       granite5asr IN_LIST AUDIOCPP_LINKED_MODELS)
        add_executable(nemo_mel_migration_probe tests/nemo_mel_migration/probe.cpp)
        target_link_libraries(nemo_mel_migration_probe PRIVATE engine_runtime ggml)
    endif()

    if(index_tts2 IN_LIST AUDIOCPP_LINKED_MODELS AND
       confucius4_tts IN_LIST AUDIOCPP_LINKED_MODELS AND
       seed_vc IN_LIST AUDIOCPP_LINKED_MODELS AND
       glm_tts IN_LIST AUDIOCPP_LINKED_MODELS AND
       niagara_asr IN_LIST AUDIOCPP_LINKED_MODELS AND
       qwen3_tts IN_LIST AUDIOCPP_LINKED_MODELS AND
       vieneu_v3_turbo IN_LIST AUDIOCPP_LINKED_MODELS AND
       dramabox IN_LIST AUDIOCPP_LINKED_MODELS AND
       mira_tts IN_LIST AUDIOCPP_LINKED_MODELS AND
       chatterbox IN_LIST AUDIOCPP_LINKED_MODELS AND
       dots_tts IN_LIST AUDIOCPP_LINKED_MODELS AND
       kroko_asr IN_LIST AUDIOCPP_LINKED_MODELS)
        add_executable(shared_mel_migration_probe tests/shared_mel_migration/probe.cpp)
        target_link_libraries(shared_mel_migration_probe PRIVATE engine_runtime ggml)
        target_include_directories(shared_mel_migration_probe PRIVATE src/models/chatterbox)
    endif()

    set(ENGINE_UNITTEST_ASSET_ROOT "${CMAKE_CURRENT_SOURCE_DIR}/tests/unittests/assets")

    function(add_engine_unittest target source)
        add_executable(${target} ${source})
        target_link_libraries(${target} PRIVATE engine_runtime ggml)
    endfunction()

    if (whistle_asr IN_LIST AUDIOCPP_LINKED_MODELS)
        add_executable(whistle_frontend_test
            tests/whistle_asr/test_frontend.cpp
            src/community_models/whistle_asr/frontend.cpp)
        target_link_libraries(whistle_frontend_test PRIVATE engine_runtime ggml)
        add_test(NAME whistle_frontend_test COMMAND whistle_frontend_test)

        add_executable(whistle_tokenizer_test
            tests/whistle_asr/test_tokenizer.cpp
            src/community_models/whistle_asr/assets.cpp)
        target_link_libraries(whistle_tokenizer_test PRIVATE engine_runtime ggml)
        add_test(NAME whistle_tokenizer_test COMMAND whistle_tokenizer_test)

        add_executable(whistle_assets_test
            tests/whistle_asr/test_assets.cpp
            src/community_models/whistle_asr/assets.cpp
            src/community_models/whistle_asr/decoder.cpp
            src/community_models/whistle_asr/encoder.cpp
            src/community_models/whistle_asr/frontend.cpp
            src/community_models/whistle_asr/graph.cpp
            src/community_models/whistle_asr/runtime.cpp)
        target_link_libraries(whistle_assets_test PRIVATE engine_runtime ggml)
        if (ENGINE_ENABLE_OPENMP)
            target_link_libraries(whistle_assets_test PRIVATE OpenMP::OpenMP_CXX)
        endif()
        set(AUDIOCPP_WHISTLE_TEST_MODEL "" CACHE FILEPATH "Locally converted Whistle GGUF for integration tests")
        if (AUDIOCPP_WHISTLE_TEST_MODEL)
            add_test(NAME whistle_asr_integration COMMAND whistle_assets_test
                "${AUDIOCPP_WHISTLE_TEST_MODEL}"
                "${CMAKE_CURRENT_SOURCE_DIR}/assets/resources/sample_16k.wav"
                "Some call me Nature.")
            add_test(NAME whistle_asr_full_integration COMMAND whistle_assets_test
                "${AUDIOCPP_WHISTLE_TEST_MODEL}"
                "${CMAKE_CURRENT_SOURCE_DIR}/assets/resources/sample_16k.wav"
                "Some call me Nature. Others call me Mother Nature. I've been here for over 4.5 billion years, twenty two thousand five hundred times longer than you."
                --full)
        endif()
    endif()

    add_engine_unittest(shared_weight_cache_test tests/unittests/test_shared_weight_cache.cpp)
    add_test(NAME shared_weight_cache_test COMMAND shared_weight_cache_test)
    add_engine_unittest(cpu_extra_buffers_test tests/unittests/test_cpu_extra_buffers.cpp)
    add_test(NAME cpu_extra_buffers_test COMMAND cpu_extra_buffers_test)
    add_engine_unittest(rknn_engine_test tests/unittests/test_rknn_engine.cpp)
    add_test(NAME rknn_engine_test COMMAND rknn_engine_test)
    if (ENGINE_ENABLE_RKNN)
        # The driver-priced cases in the file compile only where a runtime is linked; they ask
        # the same header the engine itself compiles against.
        target_compile_definitions(rknn_engine_test PRIVATE ENGINE_HAVE_RKNN=1)
        if (ENGINE_RKNN_INCLUDE_DIR)
            target_include_directories(rknn_engine_test PRIVATE ${ENGINE_RKNN_INCLUDE_DIR})
        endif()
    endif()
    if (ENGINE_ENABLE_NEON)
        add_executable(neon_vocoder_test tests/unittests/test_neon_vocoder.cpp)
        target_link_libraries(neon_vocoder_test PRIVATE engine_neon)
        add_test(NAME neon_vocoder_test COMMAND neon_vocoder_test)
        set_tests_properties(neon_vocoder_test PROPERTIES SKIP_RETURN_CODE 77)
        add_engine_unittest(neon_blocks_test tests/unittests/test_neon_blocks.cpp)
        target_include_directories(neon_blocks_test PRIVATE src/framework/neon)
        add_test(NAME neon_blocks_test COMMAND neon_blocks_test)
        add_executable(neon_library_test tests/unittests/test_neon_library.cpp)
        target_include_directories(neon_library_test PRIVATE tests/unittests)
        target_link_libraries(neon_library_test PRIVATE engine_neon)
        add_test(NAME neon_library_test COMMAND neon_library_test)
        add_executable(neon_tensor_ops_test tests/unittests/test_neon_tensor_ops.cpp)
        target_link_libraries(neon_tensor_ops_test PRIVATE engine_neon)
        add_test(NAME neon_tensor_ops_test COMMAND neon_tensor_ops_test)
    endif()
    if (sanotts IN_LIST AUDIOCPP_LINKED_MODELS)
        add_engine_unittest(sanotts_neon_decoder_test tests/unittests/test_sanotts_neon_decoder.cpp)
        target_include_directories(sanotts_neon_decoder_test PRIVATE tests/unittests src/community_models/sanotts)
        add_test(NAME sanotts_neon_decoder_test COMMAND sanotts_neon_decoder_test)
    endif()
    if (open_wake_word IN_LIST AUDIOCPP_LINKED_MODELS)
        add_engine_unittest(open_wake_word_runtime_test tests/open_wake_word/test_runtime_parity.cpp)
        add_engine_unittest(open_wake_word_frontend_test tests/open_wake_word/test_frontend.cpp)
        set(AUDIOCPP_OPEN_WAKE_WORD_TRACE "" CACHE FILEPATH "Official OpenWakeWord parity trace GGUF")
        if (AUDIOCPP_OPEN_WAKE_WORD_TRACE)
            add_test(NAME open_wake_word_runtime_parity COMMAND open_wake_word_runtime_test
                "${AUDIOCPP_OPEN_WAKE_WORD_TRACE}" 4)
            add_test(NAME open_wake_word_frontend_parity COMMAND open_wake_word_frontend_test
                "${AUDIOCPP_OPEN_WAKE_WORD_TRACE}")
        endif()
    endif()
    if (ENGINE_ENABLE_NEON AND moonshine_asr IN_LIST AUDIOCPP_LINKED_MODELS)
        add_executable(moonshine_neon_decoder_test tests/unittests/test_moonshine_neon_decoder.cpp
            src/models/moonshine_asr/neon_decoder.cpp)
        target_include_directories(moonshine_neon_decoder_test PRIVATE include tests/unittests)
        target_link_libraries(moonshine_neon_decoder_test PRIVATE engine_neon)
        add_test(NAME moonshine_neon_decoder_test COMMAND moonshine_neon_decoder_test)
    endif()
    add_engine_unittest(recipe_test tests/unittests/test_recipe.cpp)
    target_sources(recipe_test PRIVATE ${AUDIOCPP_RECIPE_SOURCES})
    target_include_directories(recipe_test PRIVATE app/recipe)
    add_test(NAME recipe_test COMMAND recipe_test)
    if (ENGINE_ENABLE_VULKAN)
        add_engine_unittest(vulkan_shared_weight_cache_test tests/unittests/test_vulkan_shared_weight_cache.cpp)
        add_test(NAME vulkan_shared_weight_cache_test COMMAND vulkan_shared_weight_cache_test)
        set_tests_properties(vulkan_shared_weight_cache_test PROPERTIES SKIP_RETURN_CODE 77)
    endif()

    if (auk IN_LIST AUDIOCPP_LINKED_MODELS)
        add_engine_unittest(auk_conditioning_test tests/auk/conditioning_test.cpp)
        target_sources(auk_conditioning_test PRIVATE
            src/community_models/auk/audio_conditioning.cpp
            src/community_models/auk/conditioning.cpp)
        add_engine_unittest(auk_vae_test tests/auk/vae_test.cpp)
        target_sources(auk_vae_test PRIVATE src/community_models/auk/vae.cpp)
        add_engine_unittest(auk_flow_test tests/auk/flow_test.cpp)
        # The flow test drives the session end to end, so it needs every runtime the
        # session builds, not just the flow stage.
        target_sources(auk_flow_test PRIVATE
            src/community_models/auk/audio_conditioning.cpp
            src/community_models/auk/conditioning.cpp
            src/community_models/auk/flow.cpp
            src/community_models/auk/session.cpp
            src/community_models/auk/vae.cpp)
        # session.cpp includes <yaml.h>, whose include dir yaml_vendor exports PUBLIC
        # and engine_runtime does not re-export.
        target_link_libraries(auk_flow_test PRIVATE yaml_vendor)
    endif()

    if (ENGINE_ENABLE_CUDA)
        add_engine_unittest(torch_random_cuda_multi_device_test
            tests/unittests/test_torch_random_cuda_multi_device.cu)
        target_include_directories(torch_random_cuda_multi_device_test PRIVATE
            ${CMAKE_CURRENT_SOURCE_DIR}/src)
        target_link_libraries(torch_random_cuda_multi_device_test PRIVATE CUDA::cudart)
        add_test(NAME torch_random_cuda_multi_device_test COMMAND torch_random_cuda_multi_device_test)
        set_tests_properties(torch_random_cuda_multi_device_test PROPERTIES SKIP_RETURN_CODE 77)

        add_engine_unittest(cuda_bias_fusion_test tests/unittests/test_cuda_bias_fusion.cpp)
        add_test(NAME cuda_bias_fusion_test COMMAND cuda_bias_fusion_test)
        set_tests_properties(cuda_bias_fusion_test PROPERTIES SKIP_RETURN_CODE 77)
        add_engine_unittest(cuda_rope_rows_test tests/unittests/test_cuda_rope_rows.cpp)
        add_test(NAME cuda_rope_rows_test COMMAND cuda_rope_rows_test)
        set_tests_properties(cuda_rope_rows_test PROPERTIES SKIP_RETURN_CODE 77)
        # This probe calls an internal CUDA symbol, not the backend registry API.
        if (NOT GGML_BACKEND_DL)
            add_engine_unittest(cuda_head_convert_test tests/unittests/test_cuda_head_convert.cpp)
            target_link_libraries(cuda_head_convert_test PRIVATE CUDA::cudart_static)
            add_test(NAME cuda_head_convert_test COMMAND cuda_head_convert_test)
            set_tests_properties(cuda_head_convert_test PROPERTIES SKIP_RETURN_CODE 77)
        endif()
    endif()

    # The v2 family is the one that builds src/community_models/sortformer_diar/.
    if (sortformer_diar_v2 IN_LIST AUDIOCPP_LINKED_MODELS)
        add_engine_unittest(sortformer_v2_aosc_test
            tests/community_models/sortformer_diar/sortformer_v2_aosc_test.cpp)
        target_sources(sortformer_v2_aosc_test PRIVATE src/community_models/sortformer_diar/aosc_state.cpp)
        add_test(NAME sortformer_v2_aosc_test COMMAND sortformer_v2_aosc_test)

        add_engine_unittest(sortformer_v2_schedule_test
            tests/community_models/sortformer_diar/sortformer_v2_schedule_test.cpp)
        target_sources(sortformer_v2_schedule_test PRIVATE
            src/community_models/sortformer_diar/frontend.cpp
            src/community_models/sortformer_diar/stream_schedule.cpp)
        add_test(NAME sortformer_v2_schedule_test COMMAND sortformer_v2_schedule_test)
    endif()

    # Links both families: the test reads the file nemotron_3_diar writes.
    if(nemotron_asr IN_LIST AUDIOCPP_LINKED_MODELS AND nemotron_3_diar IN_LIST AUDIOCPP_LINKED_MODELS)
        add_engine_unittest(nemotron_asr_speaker_tagging_test tests/nemotron_asr/speaker_tagging_test.cpp)
        add_test(NAME nemotron_asr_speaker_tagging_test COMMAND nemotron_asr_speaker_tagging_test)
    endif()

    if(nemotron_3_diar IN_LIST AUDIOCPP_LINKED_MODELS)
        add_engine_unittest(nemotron_3_diar_streaming_test tests/nemotron_3_diar/streaming_schedule_test.cpp)
        add_test(NAME nemotron_3_diar_streaming_test COMMAND nemotron_3_diar_streaming_test)
    endif()

    if(yue2 IN_LIST AUDIOCPP_LINKED_MODELS)
        add_engine_unittest(yue2_request_test tests/yue2/request_test.cpp)
        add_test(NAME yue2_request_test COMMAND yue2_request_test)
    endif()

    # Registered here rather than in the extended block below so it runs in the
    # linux/mac build workflows (they only set ENGINE_BUILD_TESTS). It is
    # platform-safe: the Metal timing section skips itself when Metal is absent.
    add_engine_unittest(conv1d_pertap_fast_path_test tests/unittests/test_conv1d_pertap_fast_path.cpp)
    add_test(NAME conv1d_pertap_fast_path_test COMMAND conv1d_pertap_fast_path_test)
    add_engine_unittest(kaldi_fbank_sparse_test tests/unittests/test_kaldi_fbank_sparse.cpp)
    add_test(NAME kaldi_fbank_sparse_test COMMAND kaldi_fbank_sparse_test)

    if(audio8_tts IN_LIST AUDIOCPP_LINKED_MODELS)
        # Regression test for the Falcon-H1 host KV cache append (per-head re-stride).
        add_engine_unittest(audio8_tts_falcon_kv_cache_test tests/unittests/test_audio8_tts_falcon_kv_cache.cpp)
        target_include_directories(audio8_tts_falcon_kv_cache_test PRIVATE
            ${CMAKE_CURRENT_SOURCE_DIR}/src/community_models/audio8_tts)
        add_test(NAME audio8_tts_falcon_kv_cache_test COMMAND audio8_tts_falcon_kv_cache_test)
    endif()

    add_engine_unittest(phonon_hotwords_test tests/parakeet_tdt/test_hotwords.cpp)
    target_sources(phonon_hotwords_test PRIVATE src/community_models/parakeet_tdt/hotwords.cpp)
    target_include_directories(phonon_hotwords_test PRIVATE ${CMAKE_CURRENT_SOURCE_DIR}/external/llama_tokenizer)
    add_test(NAME phonon_hotwords_test COMMAND phonon_hotwords_test)

    add_engine_unittest(parakeet_variant_config_test tests/parakeet_tdt/test_variant_config.cpp)
    target_sources(parakeet_variant_config_test PRIVATE src/community_models/parakeet_tdt/assets.cpp)
    target_include_directories(parakeet_variant_config_test PRIVATE ${CMAKE_CURRENT_SOURCE_DIR}/tests/unittests)
    target_compile_definitions(parakeet_variant_config_test PRIVATE ENGINE_REPO_ROOT="${CMAKE_CURRENT_SOURCE_DIR}")
    add_test(NAME parakeet_variant_config_test COMMAND parakeet_variant_config_test)

    # The probe drives the registry/session path end to end, so it needs the
    # full parakeet_tdt model library.
    if (parakeet_tdt IN_LIST AUDIOCPP_LINKED_MODELS)
        add_engine_unittest(phonon_features_probe tests/parakeet_tdt/phonon_features_probe.cpp)
    endif()

    if (granite5asr IN_LIST AUDIOCPP_LINKED_MODELS)
        add_engine_unittest(granite5asr_ctc_timestamps_test tests/granite5asr/test_ctc_timestamps.cpp)
        target_include_directories(granite5asr_ctc_timestamps_test PRIVATE ${CMAKE_CURRENT_SOURCE_DIR}/tests/unittests)
        add_test(NAME granite5asr_ctc_timestamps_test COMMAND granite5asr_ctc_timestamps_test)
    endif()

    if (ENGINE_BUILD_EXTENDED_TESTS AND ENGINE_ENABLE_METAL)
        add_executable(metal_im2col_test tests/unittests/test_metal_im2col.cpp)
        target_link_libraries(metal_im2col_test PRIVATE ggml ggml-metal)
        target_compile_definitions(metal_im2col_test PRIVATE
            METAL_IM2COL_GOLDEN_PATH="${ENGINE_UNITTEST_ASSET_ROOT}/metal_im2col/original_metal_sha256.txt")
        add_test(NAME metal_im2col_test COMMAND metal_im2col_test)
        set_tests_properties(metal_im2col_test PROPERTIES
            SKIP_RETURN_CODE 77 LABELS "extended;metal")
    endif()

    if (ENGINE_BUILD_TESTS)
        # Runtime tests exercise the core HTTP handlers, not optional frontend
        # modules. Give their frontend.cpp an empty registry and exclude the
        # server target's inherited frontend object files.
        set(AUDIOCPP_SERVER_TEST_GENERATED_DIR
            "${CMAKE_CURRENT_BINARY_DIR}/generated/server_runtime_tests")
        file(MAKE_DIRECTORY "${AUDIOCPP_SERVER_TEST_GENERATED_DIR}")
        file(WRITE "${AUDIOCPP_SERVER_TEST_GENERATED_DIR}/server_frontend_module_declarations.inc" "")
        file(WRITE "${AUDIOCPP_SERVER_TEST_GENERATED_DIR}/server_frontend_module_registrations.inc" "")

        add_executable(server_speech_option_exceptions_test
            tests/unittests/test_server_speech_option_exceptions.cpp)
        target_include_directories(server_speech_option_exceptions_test PRIVATE ${CMAKE_CURRENT_SOURCE_DIR}/app/server)
        add_test(NAME server_speech_option_exceptions_test COMMAND server_speech_option_exceptions_test)

        add_executable(server_model_slots_test tests/unittests/test_server_model_slots.cpp)
        target_include_directories(server_model_slots_test PRIVATE ${CMAKE_CURRENT_SOURCE_DIR}/app/server ${CMAKE_CURRENT_SOURCE_DIR}/include)
        target_link_libraries(server_model_slots_test PRIVATE Threads::Threads)
        add_test(NAME server_model_slots_test COMMAND server_model_slots_test)
        set_tests_properties(server_model_slots_test PROPERTIES TIMEOUT 60)

        get_target_property(server_selection_sources audiocpp_server SOURCES)
        list(FILTER server_selection_sources EXCLUDE REGEX "app/server/(main|model_installer|parallel_runtime)\\.cpp$")
        list(FILTER server_selection_sources EXCLUDE REGEX "audiocpp_server_frontends")
        add_executable(server_runtime_selection_test
            tests/unittests/test_server_runtime_selection.cpp ${server_selection_sources})
        target_include_directories(server_runtime_selection_test PRIVATE
            ${CMAKE_CURRENT_SOURCE_DIR}/app/server
            "${AUDIOCPP_SERVER_TEST_GENERATED_DIR}"
            "${CMAKE_CURRENT_BINARY_DIR}/generated")
        target_link_libraries(server_runtime_selection_test PRIVATE engine_runtime ggml Threads::Threads)
        if(WIN32)
            target_link_libraries(server_runtime_selection_test PRIVATE ws2_32)
        endif()
        add_test(NAME server_runtime_selection_test COMMAND server_runtime_selection_test)
        set_tests_properties(server_runtime_selection_test PROPERTIES TIMEOUT 60)

        add_executable(server_voice_preset_test
            tests/unittests/test_server_voice_preset.cpp ${server_selection_sources})
        target_include_directories(server_voice_preset_test PRIVATE
            ${CMAKE_CURRENT_SOURCE_DIR}/app/server
            "${AUDIOCPP_SERVER_TEST_GENERATED_DIR}"
            "${CMAKE_CURRENT_BINARY_DIR}/generated")
        target_link_libraries(server_voice_preset_test PRIVATE engine_runtime ggml Threads::Threads)
        if(WIN32)
            target_link_libraries(server_voice_preset_test PRIVATE ws2_32)
        endif()
        add_test(NAME server_voice_preset_test COMMAND server_voice_preset_test)

        # Exercise real handlers with controlled, tiny model loaders. No native
        # installer or backend-specific model admission is needed by this test.
        get_target_property(server_lifecycle_sources audiocpp_server SOURCES)
        list(FILTER server_lifecycle_sources EXCLUDE REGEX "app/server/(main|model_installer|runtime)\\.cpp$")
        list(FILTER server_lifecycle_sources EXCLUDE REGEX "audiocpp_server_frontends")
        add_executable(server_parallel_lifecycle_test
            tests/unittests/test_server_parallel_lifecycle.cpp ${server_lifecycle_sources})
        target_include_directories(server_parallel_lifecycle_test PRIVATE
            ${CMAKE_CURRENT_SOURCE_DIR}/app/server
            "${AUDIOCPP_SERVER_TEST_GENERATED_DIR}"
            "${CMAKE_CURRENT_BINARY_DIR}/generated")
        target_link_libraries(server_parallel_lifecycle_test PRIVATE engine_runtime ggml Threads::Threads)
        if(WIN32)
            target_link_libraries(server_parallel_lifecycle_test PRIVATE ws2_32)
        endif()
        add_test(NAME server_parallel_lifecycle_test COMMAND server_parallel_lifecycle_test)
        set_tests_properties(server_parallel_lifecycle_test PROPERTIES TIMEOUT 90)

        add_executable(session_pool_test tests/unittests/test_session_pool.cpp)
        target_include_directories(session_pool_test PRIVATE ${CMAKE_CURRENT_SOURCE_DIR}/include ${CMAKE_CURRENT_SOURCE_DIR}/app/server)
        # TaskResult owns JSON values for custom-schema outputs.
        target_link_libraries(session_pool_test PRIVATE engine_runtime ggml-base Threads::Threads)
        add_test(NAME session_pool_test COMMAND session_pool_test)
        set_tests_properties(session_pool_test PROPERTIES TIMEOUT 60)

        if(kokoro_tts IN_LIST AUDIOCPP_LINKED_MODELS)
            add_engine_unittest(kokoro_cpu_kernel_test tests/kokoro_tts/kokoro_cpu_kernel_test.cpp)
            add_engine_unittest(kokoro_g2p_probe tests/kokoro_tts/kokoro_g2p_probe.cpp)
            add_test(NAME kokoro_cpu_kernel_test COMMAND kokoro_cpu_kernel_test)
            if(AUDIOCPP_STATIC_ESPEAK)
                audiocpp_stage_espeak(kokoro_g2p_probe)
            endif()
        endif()
        add_engine_unittest(espeak_data_test tests/unittests/test_espeak_data.cpp)
        target_link_libraries(espeak_data_test PRIVATE Threads::Threads)
        add_test(NAME espeak_data_test COMMAND espeak_data_test)
        add_library(espeak_test_library SHARED tests/unittests/espeak_test_library.cpp)
        add_library(espeak_test_missing_symbol SHARED tests/unittests/espeak_test_library.cpp)
        target_link_libraries(espeak_test_library PRIVATE Threads::Threads)
        target_link_libraries(espeak_test_missing_symbol PRIVATE Threads::Threads)
        target_compile_definitions(espeak_test_missing_symbol PRIVATE ESPEAK_TEST_MISSING_SYMBOL=1)
        add_engine_unittest(espeak_phonemizer_test tests/unittests/test_espeak_phonemizer.cpp)
        target_link_libraries(espeak_phonemizer_test PRIVATE Threads::Threads)
        add_dependencies(espeak_phonemizer_test espeak_test_library espeak_test_missing_symbol)
        add_test(NAME espeak_phonemizer_test COMMAND espeak_phonemizer_test
            $<TARGET_FILE:espeak_test_library> $<TARGET_FILE:espeak_test_missing_symbol>)
        add_executable(sentencepiece_tokenizer1_test
            tests/unittests/test_sentencepiece_tokenizer1.cpp
        )

        target_link_libraries(sentencepiece_tokenizer1_test PRIVATE engine_runtime ggml)
        target_compile_definitions(sentencepiece_tokenizer1_test PRIVATE
            ENGINE_TEST_ASSET_ROOT="${ENGINE_UNITTEST_ASSET_ROOT}"
        )

        add_test(
            NAME sentencepiece_tokenizer1_test
            COMMAND sentencepiece_tokenizer1_test
        )

        add_engine_unittest(citrinet_tokenizer_decode_test tests/unittests/test_citrinet_tokenizer_decode.cpp)
        target_compile_definitions(citrinet_tokenizer_decode_test PRIVATE
            ENGINE_TEST_ASSET_ROOT="${ENGINE_UNITTEST_ASSET_ROOT}"
        )

        add_test(
            NAME citrinet_tokenizer_decode_test
            COMMAND citrinet_tokenizer_decode_test
        )

        add_engine_unittest(audio_dsp_test tests/unittests/test_audio_dsp.cpp)
        add_test(NAME audio_dsp_test COMMAND audio_dsp_test)

        add_engine_unittest(mel_spectrogram_frontend_test
            tests/unittests/test_mel_spectrogram_frontend.cpp)
        add_test(NAME mel_spectrogram_frontend_test COMMAND mel_spectrogram_frontend_test)

        if(sheetsage2 IN_LIST AUDIOCPP_LINKED_MODELS)
            add_engine_unittest(sheetsage_audio_frontend_test tests/unittests/test_sheetsage_audio_frontend.cpp)
            add_test(NAME sheetsage_audio_frontend_test COMMAND sheetsage_audio_frontend_test --log)
            add_engine_unittest(sheetsage_processing_test tests/unittests/test_sheetsage_processing.cpp)
            add_test(NAME sheetsage_processing_test COMMAND sheetsage_processing_test --log)
        endif()

        if(lfm2_audio IN_LIST AUDIOCPP_LINKED_MODELS)
            foreach(lfm2_audio_test IN ITEMS
                lfm2_audio_asr_inputs_test
                lfm2_audio_assets_test
                lfm2_audio_backbone_test
                lfm2_audio_chat_test
                lfm2_audio_encoder_test
                lfm2_audio_features_test
                lfm2_audio_session_test
                lfm2_audio_tokenizer_test
                lfm2_audio_tts_test)
                add_engine_unittest(${lfm2_audio_test} tests/community_models/lfm2_audio/${lfm2_audio_test}.cpp)
                target_include_directories(${lfm2_audio_test} PRIVATE ${CMAKE_CURRENT_SOURCE_DIR}/tests/unittests)
                target_compile_definitions(${lfm2_audio_test} PRIVATE ENGINE_REPO_ROOT="${CMAKE_CURRENT_SOURCE_DIR}")
                add_test(NAME ${lfm2_audio_test} COMMAND ${lfm2_audio_test})
            endforeach()

            if (ENGINE_BUILD_EXTENDED_TESTS)
                # Keep AMX repacking coverage outside normal CI until its single-block bug is fixed.
                add_test(NAME lfm2_audio_encoder_cpu_repack_test
                    COMMAND lfm2_audio_encoder_test --cpu-repack-only)
            endif()

            if (ENGINE_ENABLE_CUDA)
                add_engine_unittest(lfm2_audio_cuda_test tests/community_models/lfm2_audio/lfm2_audio_cuda_test.cpp)
                target_include_directories(lfm2_audio_cuda_test PRIVATE ${CMAKE_CURRENT_SOURCE_DIR}/tests/unittests)
                add_test(NAME lfm2_audio_cuda_test COMMAND lfm2_audio_cuda_test)
                set_tests_properties(lfm2_audio_cuda_test PROPERTIES SKIP_RETURN_CODE 77)
            endif()
        endif()

        add_engine_unittest(midi_file_test tests/unittests/test_midi_file.cpp)
        add_test(NAME midi_file_test COMMAND midi_file_test)

        add_engine_unittest(wav_reader_test tests/unittests/test_wav_reader.cpp)
        add_test(NAME wav_reader_test COMMAND wav_reader_test)

        add_engine_unittest(wav_reader_chunk_bounds_test tests/unittests/test_wav_reader_chunk_bounds.cpp)
        add_test(NAME wav_reader_chunk_bounds_test COMMAND wav_reader_chunk_bounds_test)

        add_engine_unittest(hf_tokenizer_vocab_bounds_test tests/unittests/test_hf_tokenizer_vocab_bounds.cpp)
        add_test(NAME hf_tokenizer_vocab_bounds_test COMMAND hf_tokenizer_vocab_bounds_test)

        add_engine_unittest(hf_tokenizer_sequence_decode_test tests/unittests/test_hf_tokenizer_sequence_decode.cpp)
        add_test(NAME hf_tokenizer_sequence_decode_test COMMAND hf_tokenizer_sequence_decode_test)

        add_engine_unittest(safetensors_offsets_test tests/unittests/test_safetensors_offsets.cpp)
        add_test(NAME safetensors_offsets_test COMMAND safetensors_offsets_test)

        add_engine_unittest(safetensors_metadata_depth_test tests/unittests/test_safetensors_metadata_depth.cpp)
        add_test(NAME safetensors_metadata_depth_test COMMAND safetensors_metadata_depth_test)

        add_engine_unittest(torch_bin_storage_offset_test tests/unittests/test_torch_bin_storage_offset.cpp)
        add_test(NAME torch_bin_storage_offset_test COMMAND torch_bin_storage_offset_test)

        add_engine_unittest(asr_graph_capacity_test tests/unittests/test_asr_graph_capacity.cpp)
        add_test(NAME asr_graph_capacity_test COMMAND asr_graph_capacity_test)

        add_engine_unittest(audio_chunking_test tests/unittests/test_audio_chunking.cpp)
        add_test(NAME audio_chunking_test COMMAND audio_chunking_test)
        add_engine_unittest(partial_text_test tests/unittests/test_partial_text.cpp)
        add_test(NAME partial_text_test COMMAND partial_text_test)
        add_engine_unittest(streaming_audio_test tests/unittests/test_streaming_audio.cpp)
        add_engine_unittest(streaming_asr_session_test tests/unittests/test_streaming_asr_session.cpp)
        add_test(NAME streaming_asr_session_test COMMAND streaming_asr_session_test)
        add_test(NAME streaming_audio_test COMMAND streaming_audio_test)

        add_engine_unittest(wav_writer_formats_test tests/unittests/test_wav_writer_formats.cpp)
        add_test(NAME wav_writer_formats_test COMMAND wav_writer_formats_test)

        add_engine_unittest(chinese_normalization_test tests/unittests/test_chinese_normalization.cpp)
        add_test(NAME chinese_normalization_test COMMAND chinese_normalization_test)

        add_engine_unittest(text_chunking_test tests/unittests/test_text_chunking.cpp)
        add_test(NAME text_chunking_test COMMAND text_chunking_test)
        if (ming3_audio IN_LIST AUDIOCPP_LINKED_MODELS)
            add_executable(ming3_audio_waveform_trim_test
                tests/ming3_audio/test_ming3_audio_waveform_trim.cpp)
            target_include_directories(ming3_audio_waveform_trim_test PRIVATE
                ${CMAKE_CURRENT_SOURCE_DIR}/src/models/ming3_audio)
            target_link_libraries(ming3_audio_waveform_trim_test PRIVATE engine_runtime ggml)
            add_test(NAME ming3_audio_waveform_trim_test COMMAND ming3_audio_waveform_trim_test)
        endif()

        if (moss IN_LIST AUDIOCPP_LINKED_MODELS)
            add_executable(moss_tts_nano_text_chunking_test
                tests/moss_tts_nano/test_moss_tts_nano_text_chunking.cpp)
            target_include_directories(moss_tts_nano_text_chunking_test PRIVATE
                ${CMAKE_CURRENT_SOURCE_DIR}/src/models/moss/moss_tts_nano)
            target_link_libraries(moss_tts_nano_text_chunking_test PRIVATE engine_runtime ggml)
            add_test(NAME moss_tts_nano_text_chunking_test COMMAND moss_tts_nano_text_chunking_test)
        endif()

        # Not a model test: it exercises the framework normalizer, but through the
        # supertonic tokenizer, so in a subset set it links a model symbol.
        if (supertonic IN_LIST AUDIOCPP_LINKED_MODELS)
        add_engine_unittest(unicode_normalization_test tests/unittests/test_unicode_normalization.cpp)
        # The test exercises the framework normalizer through the supertonic
        # tokenizer, so it needs that one family source to link in a subset set.
        target_sources(unicode_normalization_test PRIVATE
            src/models/supertonic/tokenizer_text.cpp)
        add_test(NAME unicode_normalization_test COMMAND unicode_normalization_test)
        endif()

        add_engine_unittest(text_normalization_test tests/unittests/test_text_normalization.cpp)
        add_test(NAME text_normalization_test COMMAND text_normalization_test)

        if (crisperwhisper IN_LIST AUDIOCPP_LINKED_MODELS)
            add_engine_unittest(crisperwhisper_transcript_alignment_test
                tests/unittests/test_crisperwhisper_transcript_alignment.cpp)
            add_test(NAME crisperwhisper_transcript_alignment_test
                COMMAND crisperwhisper_transcript_alignment_test)
        endif()

        add_engine_unittest(subtitle_formatter_test tests/unittests/test_subtitle_formatter.cpp)
        add_test(NAME subtitle_formatter_test COMMAND subtitle_formatter_test)

        add_engine_unittest(torch_random_test tests/unittests/test_torch_random.cpp)
        add_test(NAME torch_random_test COMMAND torch_random_test)

        add_engine_unittest(backend_device_resolution_test tests/unittests/test_backend_device_resolution.cpp)
        target_include_directories(backend_device_resolution_test PRIVATE ${CMAKE_CURRENT_SOURCE_DIR}/tests/unittests)
        add_test(NAME backend_device_resolution_test COMMAND backend_device_resolution_test)

        add_engine_unittest(attention_fallback_test tests/unittests/test_attention_fallback.cpp)
        target_include_directories(attention_fallback_test PRIVATE ${CMAKE_CURRENT_SOURCE_DIR}/tests/unittests)
        add_test(NAME attention_fallback_test COMMAND attention_fallback_test)

        add_engine_unittest(partial_text_render_test tests/unittests/test_partial_text_render.cpp)
        add_test(NAME partial_text_render_test COMMAND partial_text_render_test)

        add_engine_unittest(hf_sampler_test tests/unittests/test_hf_sampler.cpp)
        add_test(NAME hf_sampler_test COMMAND hf_sampler_test)

        add_engine_unittest(diffusion_math_test tests/unittests/test_diffusion_math.cpp)
        add_test(NAME diffusion_math_test COMMAND diffusion_math_test)

        add_engine_unittest(encoder_module_test tests/unittests/test_encoder_modules.cpp)
        add_test(NAME encoder_module_test COMMAND encoder_module_test)

        add_engine_unittest(snake_beta_test tests/unittests/test_snake_beta.cpp)
        add_test(NAME snake_beta_test COMMAND snake_beta_test --log)

        add_engine_unittest(disk_expert_weight_store_test tests/unittests/test_disk_expert_weight_store.cpp)
        if (ENGINE_ENABLE_OPENMP)
            target_link_libraries(disk_expert_weight_store_test PRIVATE OpenMP::OpenMP_CXX)
        endif()
        add_test(NAME disk_expert_weight_store_test COMMAND disk_expert_weight_store_test --log)

        add_engine_unittest(transpose_module_test tests/unittests/test_transpose_module.cpp)
        add_engine_unittest(lstm_sequence_test tests/unittests/test_lstm_sequence.cpp)
        add_test(NAME lstm_sequence_test COMMAND lstm_sequence_test --log)
        target_include_directories(transpose_module_test PRIVATE ${CMAKE_CURRENT_SOURCE_DIR}/tests/unittests)
        add_test(NAME transpose_module_test COMMAND transpose_module_test)

        if (fish_drama3 IN_LIST AUDIOCPP_LINKED_MODELS)
            add_engine_unittest(fish_drama3_sampler_test tests/community_models/fish_drama3/sampler_test.cpp)
            add_test(NAME fish_drama3_sampler_test COMMAND fish_drama3_sampler_test --log)
            add_engine_unittest(fish_drama3_backbone_allocation_test tests/community_models/fish_drama3/backbone_allocation_test.cpp)
            add_test(NAME fish_drama3_backbone_allocation_test COMMAND fish_drama3_backbone_allocation_test)
            add_engine_unittest(fish_drama3_codec_source_test tests/community_models/fish_drama3/codec_source_test.cpp)
            add_test(NAME fish_drama3_codec_source_test COMMAND fish_drama3_codec_source_test)
            add_engine_unittest(fish_drama3_expert_quantization_test tests/community_models/fish_drama3/expert_quantization_test.cpp)
            add_test(NAME fish_drama3_expert_quantization_test COMMAND fish_drama3_expert_quantization_test)
            add_engine_unittest(fish_drama3_tokenizer_test tests/community_models/fish_drama3/tokenizer_test.cpp)
            set(AUDIOCPP_DRAMA3_TOKENIZER_DIR "" CACHE PATH "Optional Drama 3 tokenizer sidecars for reference tests")
            if(AUDIOCPP_DRAMA3_TOKENIZER_DIR)
                add_test(NAME fish_drama3_tokenizer_test COMMAND fish_drama3_tokenizer_test
                    "${AUDIOCPP_DRAMA3_TOKENIZER_DIR}"
                    "${CMAKE_CURRENT_SOURCE_DIR}/tests/community_models/fish_drama3/tokenizer_vectors.json")
            endif()
        endif()

        add_engine_unittest(gguf_tensor_source_test tests/unittests/test_gguf_tensor_source.cpp)
        target_include_directories(gguf_tensor_source_test PRIVATE ${CMAKE_CURRENT_SOURCE_DIR}/tests/unittests)
        add_test(NAME gguf_tensor_source_test COMMAND gguf_tensor_source_test)
        add_engine_unittest(weight_adapter_source_test tests/unittests/test_weight_adapter_source.cpp)
        add_test(NAME weight_adapter_source_test COMMAND weight_adapter_source_test --log)
        add_engine_unittest(weight_adapter_test tests/unittests/test_weight_adapter.cpp)
        add_test(NAME weight_adapter_test COMMAND weight_adapter_test --log)

        add_engine_unittest(transformer_kv_ring_test tests/unittests/test_transformer_kv_ring.cpp)
        add_test(NAME transformer_kv_ring_test COMMAND transformer_kv_ring_test)

        add_engine_unittest(kokoro_word_timings_test tests/unittests/test_kokoro_word_timings.cpp)
        target_include_directories(kokoro_word_timings_test PRIVATE ${CMAKE_CURRENT_SOURCE_DIR}/tests/unittests)
        target_link_libraries(kokoro_word_timings_test PRIVATE engine_model_kokoro_tts)
        add_test(NAME kokoro_word_timings_test COMMAND kokoro_word_timings_test)

        if (kokoro_tts IN_LIST AUDIOCPP_LINKED_MODELS)
        add_engine_unittest(kokoro_voice_blending_test tests/unittests/test_kokoro_voice_blending.cpp)
        add_test(NAME kokoro_voice_blending_test COMMAND kokoro_voice_blending_test)
        endif()

        add_engine_unittest(quantize_chunking_test tests/unittests/test_quantize_chunking.cpp)
        add_test(NAME quantize_chunking_test COMMAND quantize_chunking_test)

        add_engine_unittest(tq2_1_test tests/unittests/test_tq2_1.cpp)
        target_include_directories(tq2_1_test PRIVATE external/ggml/src)
        add_test(NAME tq2_1_test COMMAND tq2_1_test)

        add_engine_unittest(moss_ttsd_speakers_test tests/unittests/test_moss_ttsd_speakers.cpp)
        target_link_libraries(moss_ttsd_speakers_test PRIVATE engine_model_moss_ttsd)
        add_test(NAME moss_ttsd_speakers_test COMMAND moss_ttsd_speakers_test)

        add_engine_unittest(model_spec_system_test tests/unittests/test_model_spec_system.cpp)
        # So the test can read the specs this repository ships, rather than a
        # copy of them staged somewhere.
        target_compile_definitions(model_spec_system_test
            PRIVATE AUDIOCPP_SOURCE_DIR="${CMAKE_CURRENT_SOURCE_DIR}")
        target_include_directories(model_spec_system_test PRIVATE ${CMAKE_CURRENT_SOURCE_DIR}/tests/unittests)
        add_test(NAME model_spec_system_test COMMAND model_spec_system_test)
    endif()

        add_executable(moss_tts_v15_text_normalization_test
            tests/moss_tts_v15/text_normalization_test.cpp
        )
        target_link_libraries(moss_tts_v15_text_normalization_test PRIVATE engine_runtime ggml)
        add_test(NAME moss_tts_v15_text_normalization
            COMMAND moss_tts_v15_text_normalization_test --vectors
                "${CMAKE_CURRENT_SOURCE_DIR}/tests/moss_tts_v15/reference/text_normalization_vectors.json")
        # Kept separate from the shipped vectors so upstream's set stays as it
        # ships: these are cases it does not reach, with expectations taken from
        # the reference implementation rather than written by hand.
        add_test(NAME moss_tts_v15_text_normalization_regressions
            COMMAND moss_tts_v15_text_normalization_test --vectors
                "${CMAKE_CURRENT_SOURCE_DIR}/tests/moss_tts_v15/reference/text_normalization_regressions.json")

    if (ENGINE_BUILD_EXTENDED_TESTS)
        find_package(Threads REQUIRED)

        add_test(NAME audiocpp_cli_list_devices COMMAND audiocpp_cli --list-devices)
        add_test(NAME audiocpp_server_list_devices COMMAND audiocpp_server --list-devices)
        set_tests_properties(
            audiocpp_cli_list_devices
            audiocpp_server_list_devices
            PROPERTIES PASS_REGULAR_EXPRESSION "available_devices=[0-9]+"
        )

        add_engine_unittest(rnnoise_utility_test tests/unittests/test_rnnoise_utility.cpp)
        target_compile_definitions(rnnoise_utility_test PRIVATE
            ENGINE_REPO_ROOT="${CMAKE_CURRENT_SOURCE_DIR}"
            ENGINE_TEST_ASSET_ROOT="${ENGINE_UNITTEST_ASSET_ROOT}"
        )
        add_test(NAME rnnoise_utility_test COMMAND rnnoise_utility_test)

        add_engine_unittest(flashsr_utility_test tests/unittests/test_flashsr_utility.cpp)
        target_compile_definitions(flashsr_utility_test PRIVATE
            ENGINE_REPO_ROOT="${CMAKE_CURRENT_SOURCE_DIR}"
            ENGINE_TEST_ASSET_ROOT="${ENGINE_UNITTEST_ASSET_ROOT}"
        )
        add_test(NAME flashsr_utility_test COMMAND flashsr_utility_test)

        add_engine_unittest(zipenhancer_utility_test tests/unittests/test_zipenhancer_utility.cpp)
        target_compile_definitions(zipenhancer_utility_test PRIVATE
            ENGINE_REPO_ROOT="${CMAKE_CURRENT_SOURCE_DIR}"
            ENGINE_TEST_ASSET_ROOT="${ENGINE_UNITTEST_ASSET_ROOT}"
        )
        add_test(NAME zipenhancer_utility_test COMMAND zipenhancer_utility_test)

        add_engine_unittest(gtcrn_utility_test tests/unittests/test_gtcrn_utility.cpp)
        target_compile_definitions(gtcrn_utility_test PRIVATE
            ENGINE_REPO_ROOT="${CMAKE_CURRENT_SOURCE_DIR}"
            ENGINE_TEST_ASSET_ROOT="${ENGINE_UNITTEST_ASSET_ROOT}"
        )
        add_test(NAME gtcrn_utility_test COMMAND gtcrn_utility_test)

        add_engine_unittest(audio_utility_api_test tests/unittests/test_audio_utility_api.cpp)
        target_compile_definitions(audio_utility_api_test PRIVATE
            ENGINE_REPO_ROOT="${CMAKE_CURRENT_SOURCE_DIR}"
        )
        add_test(NAME audio_utility_api_test COMMAND audio_utility_api_test)

        add_executable(streaming_audio_input_test
            tests/unittests/test_streaming_audio_input.cpp
            app/streaming/pcm_source.cpp
            app/streaming/streaming.cpp
        )
        target_link_libraries(streaming_audio_input_test PRIVATE engine_runtime ggml)
        target_include_directories(streaming_audio_input_test PRIVATE ${CMAKE_CURRENT_SOURCE_DIR}/tests/unittests)
        add_test(NAME streaming_audio_input_test COMMAND streaming_audio_input_test)

        add_executable(http_live_body_test
            tests/unittests/test_http_live_body.cpp
            app/server/http.cpp
        )
        target_link_libraries(http_live_body_test PRIVATE engine_runtime ggml)
        if(WIN32)
            target_link_libraries(http_live_body_test PRIVATE ws2_32)
        endif()
        target_include_directories(http_live_body_test PRIVATE ${CMAKE_CURRENT_SOURCE_DIR}/tests/unittests)
        add_test(NAME http_live_body_test COMMAND http_live_body_test)
        # This test talks to a real loopback socket, so a framing regression could
        # block forever rather than fail. Bound it so CI reports a failure instead.
        set_tests_properties(http_live_body_test PROPERTIES TIMEOUT 120)
        # Exercise the opt-in transport against the same framing expectations.
        add_executable(parallel_http_live_body_test
            tests/unittests/test_http_live_body.cpp app/server/http.cpp app/server/parallel_http.cpp)
        target_compile_definitions(parallel_http_live_body_test PRIVATE AUDIOCPP_TEST_PARALLEL_HTTP=1)
        target_include_directories(parallel_http_live_body_test PRIVATE ${CMAKE_CURRENT_SOURCE_DIR}/tests/unittests)
        target_link_libraries(parallel_http_live_body_test PRIVATE engine_runtime ggml Threads::Threads)
        add_test(NAME parallel_http_live_body_test COMMAND parallel_http_live_body_test)
        set_tests_properties(http_live_body_test parallel_http_live_body_test PROPERTIES RESOURCE_LOCK http_live_body_port)
        set_tests_properties(parallel_http_live_body_test PROPERTIES TIMEOUT 120)

        add_executable(parallel_http_ownership_test
            tests/unittests/test_parallel_http_ownership.cpp app/server/http.cpp app/server/parallel_http.cpp)
        target_include_directories(parallel_http_ownership_test PRIVATE ${CMAKE_CURRENT_SOURCE_DIR}/app/server)
        target_link_libraries(parallel_http_ownership_test PRIVATE engine_runtime ggml Threads::Threads)
        add_test(NAME parallel_http_ownership_test COMMAND parallel_http_ownership_test)
        set_tests_properties(parallel_http_ownership_test PROPERTIES TIMEOUT 30)
        if(WIN32)
            target_link_libraries(parallel_http_live_body_test PRIVATE ws2_32)
            target_link_libraries(parallel_http_ownership_test PRIVATE ws2_32)
        endif()

        add_engine_unittest(conv_transpose_fast_path_test tests/unittests/test_conv_transpose_fast_path.cpp)
        add_test(NAME conv_transpose_fast_path_test COMMAND conv_transpose_fast_path_test)

        add_engine_unittest(depthwise_conv1d_lowering_test tests/unittests/test_depthwise_conv1d_lowering.cpp)
        add_test(NAME depthwise_conv1d_lowering_test COMMAND depthwise_conv1d_lowering_test)

        add_engine_unittest(conv_lowering_matrix_test tests/unittests/test_conv_lowering_matrix.cpp)
        if (ENGINE_ENABLE_HIP)
            target_compile_definitions(conv_lowering_matrix_test PRIVATE ENGINE_TEST_ENABLE_HIP=1)
        endif()
        add_test(NAME conv_lowering_matrix_test COMMAND conv_lowering_matrix_test)

        add_engine_unittest(i8_s_fused_ops_test tests/unittests/test_i8_s_fused_ops.cpp)
        add_test(NAME i8_s_fused_ops_test COMMAND i8_s_fused_ops_test)

        add_engine_unittest(i2_s_mul_mat_test tests/unittests/test_i2_s_mul_mat.cpp)
        add_test(NAME i2_s_mul_mat_test COMMAND i2_s_mul_mat_test)

        add_engine_unittest(scaled_dot_product_attention_test tests/unittests/test_scaled_dot_product_attention.cpp)
        add_test(NAME scaled_dot_product_attention_test COMMAND scaled_dot_product_attention_test)

        add_executable(cli_request_options_test
            tests/unittests/test_cli_request_options.cpp
            app/cli/args.cpp
            app/cli/request.cpp
            app/common/base64.cpp
        )
        target_link_libraries(cli_request_options_test PRIVATE engine_runtime ggml)
        if (ENGINE_ENABLE_OPENMP)
            target_link_libraries(cli_request_options_test PRIVATE OpenMP::OpenMP_CXX)
        endif()
        add_test(NAME cli_request_options_test COMMAND cli_request_options_test)

        add_engine_unittest(file_sink_json_test tests/unittests/test_file_sink_json.cpp)
        target_sources(file_sink_json_test PRIVATE app/workflow/file_sink.cpp)
        add_test(NAME file_sink_json_test COMMAND file_sink_json_test)

        add_executable(server_multipart_test
            tests/unittests/test_server_multipart.cpp
            app/server/multipart.cpp
        )
        target_include_directories(server_multipart_test PRIVATE ${CMAKE_CURRENT_SOURCE_DIR}/app/server)
        add_test(NAME server_multipart_test COMMAND server_multipart_test)

        add_executable(base64_test
            tests/unittests/test_base64.cpp
            app/common/base64.cpp
        )
        add_test(NAME base64_test COMMAND base64_test)

        add_executable(server_busy_guard_test
            tests/unittests/test_server_busy_guard.cpp
        )
        target_include_directories(server_busy_guard_test PRIVATE ${CMAKE_CURRENT_SOURCE_DIR}/app/server)
        target_link_libraries(server_busy_guard_test PRIVATE Threads::Threads)
        add_test(NAME server_busy_guard_test COMMAND server_busy_guard_test)
        # The suite deliberately drives threads against a held lock; bound it so a
        # regression that reintroduces unbounded queuing fails instead of hanging CI.
        set_tests_properties(server_busy_guard_test PROPERTIES TIMEOUT 60)

        add_executable(server_config_test
            tests/unittests/test_server_config.cpp
            app/server/config.cpp
            app/server/model_memory.cpp
            app/cli/args.cpp
            app/cli/request.cpp
            app/common/base64.cpp
        )
        target_include_directories(server_config_test PRIVATE ${CMAKE_CURRENT_SOURCE_DIR}/app/server)
        target_link_libraries(server_config_test PRIVATE engine_runtime ggml)
        if (ENGINE_ENABLE_OPENMP)
            target_link_libraries(server_config_test PRIVATE OpenMP::OpenMP_CXX)
        endif()
        add_test(NAME server_config_test COMMAND server_config_test)

        if (AUDIOCPP_BUILD_NATIVE_MODEL_MANAGER)
            add_executable(server_model_installer_test
                tests/unittests/test_server_model_installer.cpp
                app/server/model_installer.cpp
            )
            target_include_directories(server_model_installer_test PRIVATE
                ${CMAKE_CURRENT_SOURCE_DIR}/app/server
                ${CMAKE_CURRENT_BINARY_DIR}/generated)
            target_link_libraries(server_model_installer_test PRIVATE
                Threads::Threads engine_runtime audiocpp_package_manager)
            target_compile_definitions(server_model_installer_test PRIVATE
                AUDIOCPP_NATIVE_MANAGER_FIXTURE="${CMAKE_CURRENT_SOURCE_DIR}/tests/fixtures/native_model_manager_server.py")
            add_test(NAME server_model_installer_test COMMAND server_model_installer_test)

            add_executable(package_manager_modelscope_test
                tests/unittests/test_package_manager_modelscope.cpp
            )
            target_link_libraries(package_manager_modelscope_test PRIVATE
                Threads::Threads audiocpp_package_manager)
            target_compile_definitions(package_manager_modelscope_test PRIVATE
                AUDIOCPP_NATIVE_MANAGER_FIXTURE="${CMAKE_CURRENT_SOURCE_DIR}/tests/fixtures/native_model_manager_server.py")
            add_test(NAME package_manager_modelscope_test COMMAND package_manager_modelscope_test)

            add_executable(package_manager_http_test tests/unittests/test_package_manager_http.cpp)
            target_link_libraries(package_manager_http_test PRIVATE
                Threads::Threads audiocpp_package_manager audiocpp_cpp_httplib)
            add_test(NAME package_manager_http_test COMMAND package_manager_http_test)
        endif()
    endif()

    if (ENGINE_BUILD_MODEL_TESTS)
        if (higgs_audio_tts IN_LIST AUDIOCPP_LINKED_MODELS)
            add_engine_unittest(higgs_audio_tts_sampler_test tests/higgs_audio_tts/test_sampler.cpp)
            add_test(NAME higgs_audio_tts_sampler_test COMMAND higgs_audio_tts_sampler_test --log)
        endif()
        if("ming3_audio" IN_LIST AUDIOCPP_LINKED_MODELS)
            add_executable(ming3_audio_thinker_parity EXCLUDE_FROM_ALL tests/ming3_audio/thinker_parity.cpp)
            target_link_libraries(ming3_audio_thinker_parity PRIVATE engine_runtime ggml)
            add_executable(ming3_audio_talker_parity EXCLUDE_FROM_ALL tests/ming3_audio/talker_parity.cpp)
            target_link_libraries(ming3_audio_talker_parity PRIVATE engine_runtime ggml)
        endif()
        if("ming3_tts" IN_LIST AUDIOCPP_LINKED_MODELS)
            add_executable(ming3_tts_component_parity EXCLUDE_FROM_ALL tests/ming3_tts/component_parity.cpp)
            target_link_libraries(ming3_tts_component_parity PRIVATE engine_runtime ggml)
        endif()
        if (ENGINE_ENABLE_RKNN)
            # The board-only half of the RKNN path: residency pricing, the driver's cached
            # self-report and a real context build. The repo ships no compiled engines, so the
            # engine travels as a configure-time path and the test self-skips without one.
            add_executable(rknn_sidecar_test tests/unittests/test_rknn_sidecar.cpp)
            target_compile_definitions(rknn_sidecar_test PRIVATE ENGINE_HAVE_RKNN=1)
            # The fake-engine section covers one sidecar per model, and sidecars are
            # compiled per linked model -- so the section compiles per linked model too,
            # or a composite without inflect_v2 (the default CI set) fails to link.
            foreach(model IN LISTS AUDIOCPP_LINKED_MODELS)
                string(TOUPPER "${model}" model_upper)
                target_compile_definitions(rknn_sidecar_test PRIVATE
                    "AUDIOCPP_TEST_HAS_MODEL_${model_upper}=1")
            endforeach()
            if (ENGINE_RKNN_INCLUDE_DIR)
                target_include_directories(rknn_sidecar_test PRIVATE ${ENGINE_RKNN_INCLUDE_DIR})
            endif()
            target_link_libraries(rknn_sidecar_test PRIVATE engine_runtime ggml)
            add_test(NAME rknn_sidecar_test COMMAND rknn_sidecar_test "${AUDIOCPP_RKNN_TEST_ENGINE}")
            set_tests_properties(rknn_sidecar_test PROPERTIES SKIP_RETURN_CODE 77)
        endif()
        if("plapre" IN_LIST AUDIOCPP_LINKED_MODELS)
            add_executable(plapre_codec_parity EXCLUDE_FROM_ALL tests/plapre/codec_parity.cpp)
            target_link_libraries(plapre_codec_parity PRIVATE engine_runtime ggml)
        endif()

        if (latentsync IN_LIST AUDIOCPP_LINKED_MODELS)
            add_executable(latentsync_graph_ops_test tests/latentsync/graph_ops.cpp)
            target_link_libraries(latentsync_graph_ops_test PRIVATE engine_runtime ggml)
            add_test(NAME latentsync_graph_ops_cpu COMMAND latentsync_graph_ops_test cpu)
            if (GGML_METAL)
                add_test(NAME latentsync_graph_ops_metal COMMAND latentsync_graph_ops_test metal)
            endif()
            add_executable(latentsync_component_parity tests/latentsync/component_parity.cpp)
            target_link_libraries(latentsync_component_parity PRIVATE engine_runtime ggml)
            if (ENGINE_ENABLE_OPENMP)
                target_link_libraries(latentsync_graph_ops_test PRIVATE OpenMP::OpenMP_CXX)
                target_link_libraries(latentsync_component_parity PRIVATE OpenMP::OpenMP_CXX)
            endif()
        endif()
        if (sanotts IN_LIST AUDIOCPP_LINKED_MODELS AND inflect_v2 IN_LIST AUDIOCPP_LINKED_MODELS)
            add_engine_unittest(espeak_frontend_probe tests/unittests/espeak_frontend_probe.cpp)
            if(AUDIOCPP_STATIC_ESPEAK)
                audiocpp_stage_espeak(espeak_frontend_probe)
                add_test(NAME espeak_static_frontend_test COMMAND espeak_frontend_probe - - concurrent)
            endif()
            if (MSVC)
                target_compile_options(espeak_frontend_probe PRIVATE /utf-8)
            endif()
        endif()
        # Model-specific tests and probes.
        if (ace_step IN_LIST AUDIOCPP_LINKED_MODELS)
            add_engine_unittest(ace_step_timbre_cache_probe tests/ace_step/timbre_cache_probe.cpp)
        endif()
        if (zipvoice IN_LIST AUDIOCPP_LINKED_MODELS)
            add_executable(zipvoice_parity tests/zipvoice/zipvoice_parity_main.cpp)
            target_link_libraries(zipvoice_parity PRIVATE engine_runtime ggml)
            target_include_directories(zipvoice_parity PRIVATE ${CMAKE_CURRENT_SOURCE_DIR})
            if (ENGINE_ENABLE_OPENMP)
                target_link_libraries(zipvoice_parity PRIVATE OpenMP::OpenMP_CXX)
            endif()
            add_executable(zipvoice_zh_tokens tests/zipvoice/zipvoice_zh_tokens_main.cpp)
            target_link_libraries(zipvoice_zh_tokens PRIVATE engine_runtime)
        endif()
        if (f5_tts IN_LIST AUDIOCPP_LINKED_MODELS)
            target_compile_definitions(engine_model_f5_tts PRIVATE F5_MEL_TEST=1)
            foreach(f5_test IN ITEMS f5_e2e f5_parity f5_cfg_parity f5_tokenizer)
                add_executable(${f5_test} tests/f5_tts/${f5_test}_main.cpp)
                target_compile_definitions(${f5_test} PRIVATE F5_MEL_TEST=1)
                target_link_libraries(${f5_test} PRIVATE engine_runtime ggml)
                target_include_directories(${f5_test} PRIVATE ${CMAKE_CURRENT_SOURCE_DIR})
                if (ENGINE_ENABLE_OPENMP)
                    target_link_libraries(${f5_test} PRIVATE OpenMP::OpenMP_CXX)
                endif()
            endforeach()
        endif()

        add_executable(test_granite5asr_golden_transcription
            tests/granite5asr/test_granite5asr_golden_transcription.cpp
        )
        target_compile_definitions(test_granite5asr_golden_transcription PRIVATE
            ENGINE_REPO_ROOT="${CMAKE_CURRENT_SOURCE_DIR}"
        )
        target_link_libraries(test_granite5asr_golden_transcription PRIVATE engine_runtime ggml)
        target_include_directories(test_granite5asr_golden_transcription PRIVATE ${CMAKE_CURRENT_SOURCE_DIR})
        if (ENGINE_ENABLE_OPENMP)
            target_link_libraries(test_granite5asr_golden_transcription PRIVATE OpenMP::OpenMP_CXX)
        endif()

        if (confucius4_r2t2 IN_LIST AUDIOCPP_LINKED_MODELS)
            add_executable(test_confucius4_r2t2_graph_reuse tests/confucius4_r2t2/test_graph_reuse.cpp)
            target_compile_definitions(test_confucius4_r2t2_graph_reuse PRIVATE ENGINE_REPO_ROOT="${CMAKE_CURRENT_SOURCE_DIR}")
            target_link_libraries(test_confucius4_r2t2_graph_reuse PRIVATE engine_runtime ggml)
            add_executable(test_confucius4_r2t2_transcription
                tests/confucius4_r2t2/test_confucius4_r2t2_transcription.cpp
            )
            target_compile_definitions(test_confucius4_r2t2_transcription PRIVATE
                ENGINE_REPO_ROOT="${CMAKE_CURRENT_SOURCE_DIR}"
            )
            target_link_libraries(test_confucius4_r2t2_transcription PRIVATE engine_runtime ggml)
            target_include_directories(test_confucius4_r2t2_transcription PRIVATE ${CMAKE_CURRENT_SOURCE_DIR})
            if (ENGINE_ENABLE_OPENMP)
                target_link_libraries(test_confucius4_r2t2_transcription PRIVATE OpenMP::OpenMP_CXX)
            endif()
        endif()

        if (vibeasr IN_LIST AUDIOCPP_LINKED_MODELS)
            add_executable(test_vibeasr_vae_encoder
                tests/vibeasr/test_vibeasr_vae_encoder.cpp
            )
            target_compile_definitions(test_vibeasr_vae_encoder PRIVATE
                ENGINE_REPO_ROOT="${CMAKE_CURRENT_SOURCE_DIR}"
            )
            target_link_libraries(test_vibeasr_vae_encoder PRIVATE engine_runtime ggml)
            target_include_directories(test_vibeasr_vae_encoder PRIVATE ${CMAKE_CURRENT_SOURCE_DIR})
            if (ENGINE_ENABLE_OPENMP)
                target_link_libraries(test_vibeasr_vae_encoder PRIVATE OpenMP::OpenMP_CXX)
            endif()
            add_test(
                NAME test_vibeasr_vae_encoder
                COMMAND test_vibeasr_vae_encoder
                    --model ${CMAKE_CURRENT_SOURCE_DIR}/models/vibeasr/vibeasr-vae-encoder-i8_s.gguf
                    --audio ${CMAKE_CURRENT_SOURCE_DIR}/assets/asr_validation/librispeech/librispeech_test_clean_6930-75918-0000.wav
            )
            # Needs the converted 703 MB encoder package, which a normal checkout
            # does not have; the probe exits 125 (skip) instead of failing. Pass
            # --reference-acoustic / --reference-semantic by hand to also check
            # parity against a VibeASR.cpp dump.
            set_tests_properties(test_vibeasr_vae_encoder PROPERTIES
                SKIP_RETURN_CODE 125
                TIMEOUT 300
            )

            add_executable(test_vibeasr_asr
                tests/vibeasr/test_vibeasr_asr.cpp
            )
            target_compile_definitions(test_vibeasr_asr PRIVATE
                ENGINE_REPO_ROOT="${CMAKE_CURRENT_SOURCE_DIR}"
            )
            target_link_libraries(test_vibeasr_asr PRIVATE engine_runtime ggml)
            target_include_directories(test_vibeasr_asr PRIVATE ${CMAKE_CURRENT_SOURCE_DIR})
            if (ENGINE_ENABLE_OPENMP)
                target_link_libraries(test_vibeasr_asr PRIVATE OpenMP::OpenMP_CXX)
            endif()
            add_test(
                NAME test_vibeasr_asr
                COMMAND test_vibeasr_asr --threads 8
            )
            # Same story as the encoder probe, plus the 993 MB decoder: exits 125
            # (skip) unless both converted GGUFs sit in models/vibeasr/.
            set_tests_properties(test_vibeasr_asr PROPERTIES
                SKIP_RETURN_CODE 125
                TIMEOUT 600
            )
        endif()

        if (lfm2_audio IN_LIST AUDIOCPP_LINKED_MODELS)
            add_executable(test_lfm2_audio_asr tests/lfm2_audio/test_lfm2_audio_asr.cpp)
            target_compile_definitions(test_lfm2_audio_asr PRIVATE ENGINE_REPO_ROOT="${CMAKE_CURRENT_SOURCE_DIR}")
            target_link_libraries(test_lfm2_audio_asr PRIVATE engine_runtime ggml)
            if (ENGINE_ENABLE_OPENMP)
                target_link_libraries(test_lfm2_audio_asr PRIVATE OpenMP::OpenMP_CXX)
            endif()
            add_test(NAME test_lfm2_audio_asr COMMAND test_lfm2_audio_asr)

            # Needs the lfm2_audio_1_5b_f16 package in models/; exits 125 (skip) without it.
            set_tests_properties(test_lfm2_audio_asr PROPERTIES
                SKIP_RETURN_CODE 125
                TIMEOUT 600
            )

            add_executable(test_lfm2_audio_tts tests/lfm2_audio/test_lfm2_audio_tts.cpp)
            target_compile_definitions(test_lfm2_audio_tts PRIVATE ENGINE_REPO_ROOT="${CMAKE_CURRENT_SOURCE_DIR}")
            target_link_libraries(test_lfm2_audio_tts PRIVATE engine_runtime ggml)
            if (ENGINE_ENABLE_OPENMP)
                target_link_libraries(test_lfm2_audio_tts PRIVATE OpenMP::OpenMP_CXX)
            endif()
            add_test(NAME test_lfm2_audio_tts COMMAND test_lfm2_audio_tts)
            # Needs the lfm2_audio_1_5b_f16 package in models/; exits 125 (skip) without it.
            set_tests_properties(test_lfm2_audio_tts PROPERTIES
                SKIP_RETURN_CODE 125
                TIMEOUT 900
            )

            add_executable(test_lfm2_audio_s2s tests/lfm2_audio/test_lfm2_audio_s2s.cpp)
            target_compile_definitions(test_lfm2_audio_s2s PRIVATE ENGINE_REPO_ROOT="${CMAKE_CURRENT_SOURCE_DIR}")
            target_link_libraries(test_lfm2_audio_s2s PRIVATE engine_runtime ggml)
            if (ENGINE_ENABLE_OPENMP)
                target_link_libraries(test_lfm2_audio_s2s PRIVATE OpenMP::OpenMP_CXX)
            endif()
            add_test(NAME test_lfm2_audio_s2s COMMAND test_lfm2_audio_s2s)
            # Needs the lfm2_audio_1_5b_f16 package in models/; exits 125 (skip) without it.
            set_tests_properties(test_lfm2_audio_s2s PROPERTIES
                SKIP_RETURN_CODE 125
                TIMEOUT 900
            )
        endif()

        if (sopro_tts IN_LIST AUDIOCPP_LINKED_MODELS)
            add_executable(test_sopro_tts_audio_ops
                tests/sopro_tts/test_sopro_tts_audio_ops.cpp
            )
            target_link_libraries(test_sopro_tts_audio_ops PRIVATE engine_runtime ggml)
            target_include_directories(test_sopro_tts_audio_ops PRIVATE ${CMAKE_CURRENT_SOURCE_DIR}/tests/unittests)
            if (ENGINE_ENABLE_OPENMP)
                target_link_libraries(test_sopro_tts_audio_ops PRIVATE OpenMP::OpenMP_CXX)
            endif()

            add_test(
                NAME test_sopro_tts_audio_ops
                COMMAND test_sopro_tts_audio_ops
            )
        endif()

        if (reuse IN_LIST AUDIOCPP_LINKED_MODELS)
            add_executable(test_reuse_scan tests/reuse/test_scan.cpp app/cli/args.cpp)
            target_link_libraries(test_reuse_scan PRIVATE engine_runtime ggml)
            target_include_directories(test_reuse_scan PRIVATE ${CMAKE_CURRENT_SOURCE_DIR})
        endif()

        if (audio8_asr IN_LIST AUDIOCPP_LINKED_MODELS)
            add_executable(test_audio8_asr_units
                tests/audio8_asr/test_audio8_asr_units.cpp
            )
            target_link_libraries(test_audio8_asr_units PRIVATE engine_runtime ggml)
            target_include_directories(test_audio8_asr_units PRIVATE ${CMAKE_CURRENT_SOURCE_DIR}/tests/unittests)
            if (ENGINE_ENABLE_OPENMP)
                target_link_libraries(test_audio8_asr_units PRIVATE OpenMP::OpenMP_CXX)
            endif()

            add_test(
                NAME test_audio8_asr_units
                COMMAND test_audio8_asr_units
            )

            add_executable(test_audio8_asr_golden_transcription
                tests/audio8_asr/test_audio8_asr_golden_transcription.cpp
            )
            target_compile_definitions(test_audio8_asr_golden_transcription PRIVATE
                ENGINE_REPO_ROOT="${CMAKE_CURRENT_SOURCE_DIR}"
            )
            target_link_libraries(test_audio8_asr_golden_transcription PRIVATE engine_runtime ggml)
            target_include_directories(test_audio8_asr_golden_transcription PRIVATE ${CMAKE_CURRENT_SOURCE_DIR})
            if (ENGINE_ENABLE_OPENMP)
                target_link_libraries(test_audio8_asr_golden_transcription PRIVATE OpenMP::OpenMP_CXX)
            endif()
        endif()

        add_executable(torch_bin_parity
            tests/vibevoice/torch_bin_parity.cpp
        )
        target_link_libraries(torch_bin_parity PRIVATE engine_runtime ggml)
        if (ENGINE_ENABLE_OPENMP)
            target_link_libraries(torch_bin_parity PRIVATE OpenMP::OpenMP_CXX)
        endif()

        if (vibevoice IN_LIST AUDIOCPP_LINKED_MODELS)
            add_executable(vibevoice_finetune_overlay_check
                tests/vibevoice/finetune_overlay_check.cpp
            )
            target_link_libraries(vibevoice_finetune_overlay_check PRIVATE engine_runtime ggml)
            if (ENGINE_ENABLE_OPENMP)
                target_link_libraries(vibevoice_finetune_overlay_check PRIVATE OpenMP::OpenMP_CXX)
            endif()
        endif()

        if (moss IN_LIST AUDIOCPP_LINKED_MODELS)
        add_executable(moss_tts_local_smoke
            tests/moss_tts_local/moss_tts_local_smoke.cpp
        )
        target_link_libraries(moss_tts_local_smoke PRIVATE engine_runtime ggml)
        if (ENGINE_ENABLE_OPENMP)
            target_link_libraries(moss_tts_local_smoke PRIVATE OpenMP::OpenMP_CXX)
        endif()
        endif()

        if (moss_voicegen IN_LIST AUDIOCPP_LINKED_MODELS)
        add_executable(moss_voicegen_codec_parity
            tests/moss_voicegen/codec_decode_parity.cpp
        )
        target_link_libraries(moss_voicegen_codec_parity PRIVATE engine_runtime ggml)
        if (ENGINE_ENABLE_OPENMP)
            target_link_libraries(moss_voicegen_codec_parity PRIVATE OpenMP::OpenMP_CXX)
        endif()
        endif()

        add_executable(moss_voicegen_codec_encode_parity
            tests/moss_voicegen/codec_encode_parity.cpp
        )
        target_link_libraries(moss_voicegen_codec_encode_parity PRIVATE engine_runtime ggml)
        if (ENGINE_ENABLE_OPENMP)
            target_link_libraries(moss_voicegen_codec_encode_parity PRIVATE OpenMP::OpenMP_CXX)
        endif()

        add_executable(moss_codec_decode_length_sweep
            tests/moss_tts_delay/codec_decode_length_sweep.cpp
        )
        target_link_libraries(moss_codec_decode_length_sweep PRIVATE engine_runtime ggml)
        if (ENGINE_ENABLE_OPENMP)
            target_link_libraries(moss_codec_decode_length_sweep PRIVATE OpenMP::OpenMP_CXX)
        endif()

        add_executable(moss_tts_delay_backbone_parity
            tests/moss_tts_delay/backbone_parity.cpp
        )
        target_link_libraries(moss_tts_delay_backbone_parity PRIVATE engine_runtime ggml)
        if (ENGINE_ENABLE_OPENMP)
            target_link_libraries(moss_tts_delay_backbone_parity PRIVATE OpenMP::OpenMP_CXX)
        endif()

        # Incremental decode parity: chunked prefill / decode step / cross-utterance prefix reuse
        # against a one-shot prefill, plus frame-cache independence for the local decoder.
        if (moss IN_LIST AUDIOCPP_LINKED_MODELS)
        add_executable(moss_tts_nano_kv_cache_parity
            tests/moss_tts_nano/kv_cache_parity.cpp
        )
        target_link_libraries(moss_tts_nano_kv_cache_parity PRIVATE engine_runtime ggml)
        if (ENGINE_ENABLE_OPENMP)
            target_link_libraries(moss_tts_nano_kv_cache_parity PRIVATE OpenMP::OpenMP_CXX)
        endif()
        endif()

        # Reference-voice cache rules. Needs no downloaded model, so unlike the parity
        # target above this one is registered as a test -- but it does link the moss
        # model library for the cache helpers it exercises.
        if (moss IN_LIST AUDIOCPP_LINKED_MODELS)
        add_executable(moss_tts_nano_reference_cache_test
            tests/moss_tts_nano/reference_cache_test.cpp
        )
        target_link_libraries(moss_tts_nano_reference_cache_test PRIVATE engine_runtime ggml)
        if (ENGINE_ENABLE_OPENMP)
            target_link_libraries(moss_tts_nano_reference_cache_test PRIVATE OpenMP::OpenMP_CXX)
        endif()
        add_test(NAME moss_tts_nano_reference_cache_test COMMAND moss_tts_nano_reference_cache_test)
        endif()

        add_executable(moss_tts_v15_prompt_parity
            tests/moss_tts_v15/prompt_parity.cpp
            src/community_models/moss_tts_v15/prompt.cpp
        )
        target_link_libraries(moss_tts_v15_prompt_parity PRIVATE engine_runtime ggml)
        if (ENGINE_ENABLE_OPENMP)
            target_link_libraries(moss_tts_v15_prompt_parity PRIVATE OpenMP::OpenMP_CXX)
        endif()

        add_executable(moss_ttsd_prompt_parity
            tests/moss_ttsd/prompt_parity.cpp
            src/community_models/moss_ttsd/prompt.cpp
        )
        target_link_libraries(moss_ttsd_prompt_parity PRIVATE engine_runtime ggml)
        if (ENGINE_ENABLE_OPENMP)
            target_link_libraries(moss_ttsd_prompt_parity PRIVATE OpenMP::OpenMP_CXX)
        endif()


        if (moss_voicegen IN_LIST AUDIOCPP_LINKED_MODELS)
        add_executable(moss_voicegen_smoke
            tests/moss_voicegen/voicegen_smoke.cpp
        )
        target_link_libraries(moss_voicegen_smoke PRIVATE engine_runtime ggml)
        if (ENGINE_ENABLE_OPENMP)
            target_link_libraries(moss_voicegen_smoke PRIVATE OpenMP::OpenMP_CXX)
        endif()
        endif()

        if (moss_voicegen IN_LIST AUDIOCPP_LINKED_MODELS)
        add_executable(moss_voicegen_generation_parity
            tests/moss_voicegen/generation_parity.cpp
        )
        target_link_libraries(moss_voicegen_generation_parity PRIVATE engine_runtime ggml)
        if (ENGINE_ENABLE_OPENMP)
            target_link_libraries(moss_voicegen_generation_parity PRIVATE OpenMP::OpenMP_CXX)
        endif()
        endif()

        if (moss_voicegen IN_LIST AUDIOCPP_LINKED_MODELS)
        add_executable(moss_voicegen_backbone_parity
            tests/moss_voicegen/backbone_parity.cpp
        )
        target_link_libraries(moss_voicegen_backbone_parity PRIVATE engine_runtime ggml)
        if (ENGINE_ENABLE_OPENMP)
            target_link_libraries(moss_voicegen_backbone_parity PRIVATE OpenMP::OpenMP_CXX)
        endif()
        endif()

        if (moss_voicegen IN_LIST AUDIOCPP_LINKED_MODELS)
        add_executable(moss_voicegen_prompt_parity
            tests/moss_voicegen/prompt_parity.cpp
        )
        target_link_libraries(moss_voicegen_prompt_parity PRIVATE engine_runtime ggml)
        if (ENGINE_ENABLE_OPENMP)
            target_link_libraries(moss_voicegen_prompt_parity PRIVATE OpenMP::OpenMP_CXX)
        endif()
        endif()

        add_executable(personaplex_ring_kv_cache_parity
            tests/personaplex/personaplex_ring_kv_cache_parity.cpp
        )
        target_link_libraries(personaplex_ring_kv_cache_parity PRIVATE engine_runtime ggml)

        if (ENGINE_ENABLE_OPENMP)
            target_link_libraries(personaplex_ring_kv_cache_parity PRIVATE OpenMP::OpenMP_CXX)
        endif()

        add_executable(hifigan_vocoder_variant_parity
            tests/unittests/test_hifigan_vocoder_variants.cpp
        )
        target_link_libraries(hifigan_vocoder_variant_parity PRIVATE engine_runtime ggml)
        if (ENGINE_ENABLE_OPENMP)
            target_link_libraries(hifigan_vocoder_variant_parity PRIVATE OpenMP::OpenMP_CXX)
        endif()

        add_executable(miocodec_wavlm_parity EXCLUDE_FROM_ALL
            tests/miocodec/miocodec_wavlm_parity.cpp
        )
        target_link_libraries(miocodec_wavlm_parity PRIVATE engine_runtime ggml)

        if (dots_tts IN_LIST AUDIOCPP_LINKED_MODELS)
        add_engine_unittest(dots_tts_vocoder_parity tests/dots_tts/dots_tts_vocoder_parity.cpp)
        endif()

        # Needs the GGUF and a PyTorch reference dump, so it is driven by hand
        # rather than registered with add_test -- same as dots_tts_vocoder_parity.
        if (echo_tts IN_LIST AUDIOCPP_LINKED_MODELS)
        add_engine_unittest(echo_tts_dit_parity tests/echo_tts/echo_tts_dit_parity.cpp)
        endif()

        if (echo_tts IN_LIST AUDIOCPP_LINKED_MODELS)
        add_engine_unittest(echo_tts_host_units tests/echo_tts/echo_tts_host_units.cpp)
        target_sources(echo_tts_host_units PRIVATE
            src/community_models/echo_tts/latent_post.cpp
            src/community_models/echo_tts/tokenizer.cpp)
        add_test(
            NAME echo_tts_host_units
            COMMAND echo_tts_host_units
        )
        endif()

        if (sanotts IN_LIST AUDIOCPP_LINKED_MODELS)
        add_engine_unittest(sanotts_frontend_test tests/unittests/test_sanotts_frontend.cpp)
        target_sources(sanotts_frontend_test PRIVATE
            src/community_models/sanotts/frontend.cpp
            ${AUDIOCPP_SANOTTS_NEON_SOURCES}
            src/community_models/sanotts/runtime.cpp)
        target_include_directories(sanotts_frontend_test PRIVATE ${CMAKE_CURRENT_SOURCE_DIR}/tests/unittests)
        add_test(
            NAME sanotts_frontend_test
            COMMAND sanotts_frontend_test
        )
        endif()

        if (inflect_v2 IN_LIST AUDIOCPP_LINKED_MODELS)
        add_engine_unittest(inflect_v2_frontend_test tests/unittests/test_inflect_v2_frontend.cpp)
        target_include_directories(inflect_v2_frontend_test PRIVATE ${CMAKE_CURRENT_SOURCE_DIR}/tests/unittests)
        add_test(
            NAME inflect_v2_frontend_test
            COMMAND inflect_v2_frontend_test
        )
        endif()

        if (voxtral_realtime IN_LIST AUDIOCPP_LINKED_MODELS)
        add_engine_unittest(voxtral_realtime_stream_chunking_test tests/unittests/test_voxtral_realtime_stream_chunking.cpp)
        target_sources(voxtral_realtime_stream_chunking_test PRIVATE src/models/voxtral_realtime/frontend.cpp)
        target_include_directories(voxtral_realtime_stream_chunking_test PRIVATE ${CMAKE_CURRENT_SOURCE_DIR}/tests/unittests)
        add_test(
            NAME voxtral_realtime_stream_chunking_test
            COMMAND voxtral_realtime_stream_chunking_test
        )
        endif()

        add_engine_unittest(index_tts_tn_corpus_test tests/unittests/test_index_tts_tn_corpus.cpp)
        target_compile_definitions(index_tts_tn_corpus_test PRIVATE
            ENGINE_TEST_ASSET_ROOT="${ENGINE_UNITTEST_ASSET_ROOT}"
        )
        add_test(
            NAME index_tts_tn_corpus_test
            COMMAND index_tts_tn_corpus_test
        )

        if (outetts IN_LIST AUDIOCPP_LINKED_MODELS)
        add_engine_unittest(outetts_generation_budget_test tests/unittests/test_outetts_generation_budget.cpp)
        target_sources(outetts_generation_budget_test PRIVATE src/community_models/outetts/tokenizer.cpp)
        target_include_directories(outetts_generation_budget_test PRIVATE ${CMAKE_CURRENT_SOURCE_DIR}/tests/unittests)
        add_test(
            NAME outetts_generation_budget_test
            COMMAND outetts_generation_budget_test
        )
        endif()

        add_engine_unittest(
            decoder_packed_projection_test
            tests/unittests/test_decoder_packed_projections.cpp
        )
        add_engine_unittest(causal_decoder_chunked_prefill_test tests/unittests/test_causal_decoder_chunked_prefill.cpp)
        add_test(NAME causal_decoder_chunked_prefill_test COMMAND causal_decoder_chunked_prefill_test)
        add_test(
            NAME decoder_packed_projection_test
            COMMAND decoder_packed_projection_test
        )

        add_engine_unittest(supertonic_vector_convnext_exp_test tests/unittests/test_supertonic_vector_convnext_exp.cpp)
        add_test(
            NAME supertonic_vector_convnext_exp_test
            COMMAND supertonic_vector_convnext_exp_test
        )

        if (minimax_music3 IN_LIST AUDIOCPP_LINKED_MODELS)
        add_engine_unittest(minimax_music3_lm_head_test tests/unittests/test_minimax_music3_lm_head.cpp)
        target_sources(minimax_music3_lm_head_test PRIVATE src/community_models/minimax_music3/global_lm.cpp)
        target_include_directories(minimax_music3_lm_head_test PRIVATE ${CMAKE_CURRENT_SOURCE_DIR}/tests/unittests)
        add_test(
            NAME minimax_music3_lm_head_test
            COMMAND minimax_music3_lm_head_test
        )
        endif()

        if (minimax_music3 IN_LIST AUDIOCPP_LINKED_MODELS)
        add_engine_unittest(
            minimax_music3_pipeline_buffers_test
            tests/unittests/test_minimax_music3_pipeline_buffers.cpp)
        target_include_directories(
            minimax_music3_pipeline_buffers_test
            PRIVATE ${CMAKE_CURRENT_SOURCE_DIR}/tests/unittests)
        add_test(
            NAME minimax_music3_pipeline_buffers_test
            COMMAND minimax_music3_pipeline_buffers_test
        )
        endif()

        add_engine_unittest(
            minimax_music3_graph_release_policy_test
            tests/unittests/test_minimax_music3_graph_release_policy.cpp)
        add_test(
            NAME minimax_music3_graph_release_policy_test
            COMMAND minimax_music3_graph_release_policy_test
        )

        if (qwen3_tts IN_LIST AUDIOCPP_LINKED_MODELS)
        # Real-weight integration test; invoked explicitly with --model GGUF.
        add_engine_unittest(qwen3_tts_decoder_streaming_test tests/qwen3_tts/test_speech_decoder_streaming.cpp)
        endif()

        if (qwen3_forced_aligner IN_LIST AUDIOCPP_LINKED_MODELS)
        add_engine_unittest(qwen3_forced_aligner_processor_test tests/unittests/test_qwen3_forced_aligner_processor.cpp)
        target_sources(qwen3_forced_aligner_processor_test PRIVATE
            src/models/qwen3_asr/tokenizer_text.cpp
            src/models/qwen3_forced_aligner/processor.cpp)
        target_include_directories(qwen3_forced_aligner_processor_test PRIVATE ${CMAKE_CURRENT_SOURCE_DIR}/tests/unittests)
        add_test(
            NAME qwen3_forced_aligner_processor_test
            COMMAND qwen3_forced_aligner_processor_test
        )
        endif()

        add_engine_unittest(mms_forced_aligner_assets_test tests/unittests/test_mms_forced_aligner_assets.cpp)
        target_sources(mms_forced_aligner_assets_test PRIVATE src/community_models/mms_forced_aligner/assets.cpp)
        target_include_directories(mms_forced_aligner_assets_test PRIVATE ${CMAKE_CURRENT_SOURCE_DIR}/tests/unittests)
        add_test(
            NAME mms_forced_aligner_assets_test
            COMMAND mms_forced_aligner_assets_test
        )

        add_engine_unittest(mms_wav2vec2_positional_conv_test tests/unittests/test_mms_wav2vec2_positional_conv.cpp)
        target_include_directories(mms_wav2vec2_positional_conv_test PRIVATE ${CMAKE_CURRENT_SOURCE_DIR}/tests/unittests)
        add_test(
            NAME mms_wav2vec2_positional_conv_test
            COMMAND mms_wav2vec2_positional_conv_test
        )

        add_engine_unittest(mms_emissions_test tests/unittests/test_mms_emissions.cpp)
        target_sources(mms_emissions_test PRIVATE src/community_models/mms_forced_aligner/emissions.cpp)
        target_include_directories(mms_emissions_test PRIVATE ${CMAKE_CURRENT_SOURCE_DIR}/tests/unittests)
        add_test(
            NAME mms_emissions_test
            COMMAND mms_emissions_test
        )

        add_engine_unittest(mms_text_processor_test tests/unittests/test_mms_text_processor.cpp)
        target_sources(mms_text_processor_test PRIVATE src/community_models/mms_forced_aligner/text_processor.cpp)
        target_include_directories(mms_text_processor_test PRIVATE
            ${CMAKE_CURRENT_SOURCE_DIR}/tests/unittests
            ${CMAKE_CURRENT_SOURCE_DIR}/external/llama_tokenizer)
        add_test(
            NAME mms_text_processor_test
            COMMAND mms_text_processor_test
        )

        add_engine_unittest(mms_ctc_alignment_test tests/unittests/test_mms_ctc_alignment.cpp)
        target_sources(mms_ctc_alignment_test PRIVATE src/community_models/mms_forced_aligner/ctc_alignment.cpp)
        target_include_directories(mms_ctc_alignment_test PRIVATE ${CMAKE_CURRENT_SOURCE_DIR}/tests/unittests)
        add_test(
            NAME mms_ctc_alignment_test
            COMMAND mms_ctc_alignment_test
        )

        if (mms_forced_aligner IN_LIST AUDIOCPP_LINKED_MODELS)
        add_engine_unittest(mms_forced_aligner_session_test tests/unittests/test_mms_forced_aligner_session.cpp)
        target_sources(mms_forced_aligner_session_test PRIVATE
            src/community_models/mms_forced_aligner/assets.cpp
            src/community_models/mms_forced_aligner/ctc_alignment.cpp
            src/community_models/mms_forced_aligner/emissions.cpp
            src/community_models/mms_forced_aligner/session.cpp
            src/community_models/mms_forced_aligner/text_processor.cpp)
        target_include_directories(mms_forced_aligner_session_test PRIVATE
            ${CMAKE_CURRENT_SOURCE_DIR}/tests/unittests
            ${CMAKE_CURRENT_SOURCE_DIR}/external/llama_tokenizer)
        add_test(
            NAME mms_forced_aligner_session_test
            COMMAND mms_forced_aligner_session_test
        )
        endif()

        add_engine_unittest(soprano_text_cleaning_test tests/unittests/test_soprano_text_cleaning.cpp)
        target_sources(soprano_text_cleaning_test PRIVATE
            src/community_models/soprano_tts/text_normalizer.cpp
            src/community_models/soprano_tts/tokenizer_text.cpp)
        target_include_directories(soprano_text_cleaning_test PRIVATE ${CMAKE_CURRENT_SOURCE_DIR}/tests/unittests)
        add_test(
            NAME soprano_text_cleaning_test
            COMMAND soprano_text_cleaning_test
        )

        add_engine_unittest(fun_asr_nano_assets_test tests/unittests/test_fun_asr_nano_assets.cpp)
        target_sources(fun_asr_nano_assets_test PRIVATE src/models/fun_asr_nano/assets.cpp)
        target_include_directories(fun_asr_nano_assets_test PRIVATE ${CMAKE_CURRENT_SOURCE_DIR}/tests/unittests)
        add_test(
            NAME fun_asr_nano_assets_test
            COMMAND fun_asr_nano_assets_test
        )

        add_engine_unittest(fun_asr_nano_frontend_probe tests/fun_asr_nano/fun_asr_nano_frontend_probe.cpp)
        add_test(
            NAME fun_asr_nano_frontend_probe
            COMMAND fun_asr_nano_frontend_probe
                ${CMAKE_CURRENT_SOURCE_DIR}/tests/fun_asr_nano/frontend_reference.json
                ${CMAKE_CURRENT_SOURCE_DIR}/tests/fun_asr_nano/frontend_reference.bin
                ${CMAKE_CURRENT_SOURCE_DIR}/assets/resources/sample_16k.wav
        )

        add_engine_unittest(fun_asr_nano_sanm_probe tests/fun_asr_nano/sanm_probe.cpp)
        add_test(
            NAME fun_asr_nano_sanm_probe
            COMMAND fun_asr_nano_sanm_probe
                --backend cpu
                --reference ${CMAKE_CURRENT_SOURCE_DIR}/tests/fun_asr_nano/sanm_reference.json
                --data ${CMAKE_CURRENT_SOURCE_DIR}/tests/fun_asr_nano/sanm_reference.bin
        )

        add_engine_unittest(fun_asr_nano_encoder_probe tests/fun_asr_nano/fun_asr_nano_encoder_probe.cpp)
        target_sources(
            fun_asr_nano_encoder_probe
            PRIVATE
                src/models/fun_asr_nano/assets.cpp
                src/models/fun_asr_nano/frontend.cpp
                src/models/fun_asr_nano/encoder.cpp
        )
        set(
            AUDIOCPP_FUN_ASR_NANO_TEST_MODEL
            ""
            CACHE PATH
            "Optional local Fun-ASR-Nano model directory for full runtime parity tests"
        )
        set(
            AUDIOCPP_FUN_ASR_NANO_TEST_GGUF
            ""
            CACHE FILEPATH
            "Optional standalone Fun-ASR-Nano GGUF for full CLI transcript tests"
        )
        if (AUDIOCPP_FUN_ASR_NANO_TEST_MODEL)
            add_test(
                NAME fun_asr_nano_encoder_probe_cpu
                COMMAND fun_asr_nano_encoder_probe
                    --backend cpu
                    --model ${AUDIOCPP_FUN_ASR_NANO_TEST_MODEL}
                    --reference ${CMAKE_CURRENT_SOURCE_DIR}/tests/fun_asr_nano/encoder_reference.json
                    --data ${CMAKE_CURRENT_SOURCE_DIR}/tests/fun_asr_nano/encoder_reference.bin
            )
            if (GGML_CUDA)
                add_test(
                    NAME fun_asr_nano_encoder_probe_cuda
                    COMMAND fun_asr_nano_encoder_probe
                        --backend cuda
                        --model ${AUDIOCPP_FUN_ASR_NANO_TEST_MODEL}
                        --reference ${CMAKE_CURRENT_SOURCE_DIR}/tests/fun_asr_nano/encoder_reference.json
                        --data ${CMAKE_CURRENT_SOURCE_DIR}/tests/fun_asr_nano/encoder_reference.bin
                )
            endif()

            add_engine_unittest(fun_asr_nano_session_probe tests/fun_asr_nano/fun_asr_nano_session_probe.cpp)
            add_test(
                NAME fun_asr_nano_session_probe_cpu
                COMMAND fun_asr_nano_session_probe
                    --backend cpu
                    --model ${AUDIOCPP_FUN_ASR_NANO_TEST_MODEL}
                    --audio ${CMAKE_CURRENT_SOURCE_DIR}/assets/resources/sample_16k.wav
            )
            if (GGML_CUDA)
                add_test(
                    NAME fun_asr_nano_session_probe_cuda
                    COMMAND fun_asr_nano_session_probe
                        --backend cuda
                        --model ${AUDIOCPP_FUN_ASR_NANO_TEST_MODEL}
                        --audio ${CMAKE_CURRENT_SOURCE_DIR}/assets/resources/sample_16k.wav
                )
            endif()
        endif()
        if (AUDIOCPP_FUN_ASR_NANO_TEST_GGUF)
            add_test(
                NAME fun_asr_nano_gguf_cli_cpu
                COMMAND audiocpp_cli
                    --task asr
                    --family fun_asr_nano
                    --model ${AUDIOCPP_FUN_ASR_NANO_TEST_GGUF}
                    --backend cpu
                    --audio ${CMAKE_CURRENT_SOURCE_DIR}/assets/resources/sample_16k.wav
            )
            set_tests_properties(
                fun_asr_nano_gguf_cli_cpu
                PROPERTIES PASS_REGULAR_EXPRESSION
                    "text_output=Some call me nature, others call me mother nature"
            )
            if (GGML_CUDA)
                add_test(
                    NAME fun_asr_nano_gguf_cli_cuda
                    COMMAND audiocpp_cli
                        --task asr
                        --family fun_asr_nano
                        --model ${AUDIOCPP_FUN_ASR_NANO_TEST_GGUF}
                        --backend cuda
                        --audio ${CMAKE_CURRENT_SOURCE_DIR}/assets/resources/sample_16k.wav
                )
                set_tests_properties(
                    fun_asr_nano_gguf_cli_cuda
                    PROPERTIES PASS_REGULAR_EXPRESSION
                        "text_output=Some call me nature, others call me mother nature"
                )
                add_test(
                    NAME fun_asr_nano_gguf_cli_cuda_shared_q8
                    COMMAND audiocpp_cli
                        --task asr
                        --family fun_asr_nano
                        --model ${AUDIOCPP_FUN_ASR_NANO_TEST_GGUF}
                        --backend cuda
                        --session-option fun_asr_nano.weight_type=q8_0
                        --audio ${CMAKE_CURRENT_SOURCE_DIR}/assets/resources/sample_16k.wav
                )
                set_tests_properties(
                    fun_asr_nano_gguf_cli_cuda_shared_q8
                    PROPERTIES PASS_REGULAR_EXPRESSION
                        "text_output=Some call me nature, others call me mother nature"
                )
                add_test(
                    NAME fun_asr_nano_gguf_cli_best_shared_q8
                    COMMAND audiocpp_cli
                        --task asr
                        --family fun_asr_nano
                        --model ${AUDIOCPP_FUN_ASR_NANO_TEST_GGUF}
                        --backend best
                        --session-option fun_asr_nano.weight_type=q8_0
                        --audio ${CMAKE_CURRENT_SOURCE_DIR}/assets/resources/sample_16k.wav
                )
                set_tests_properties(
                    fun_asr_nano_gguf_cli_best_shared_q8
                    PROPERTIES PASS_REGULAR_EXPRESSION
                        "text_output=Some call me nature, others call me mother nature"
                )
            endif()
        endif()

        add_engine_unittest(fun_asr_nano_adaptor_probe tests/fun_asr_nano/fun_asr_nano_adaptor_probe.cpp)
        target_sources(
            fun_asr_nano_adaptor_probe
            PRIVATE
                src/models/fun_asr_nano/assets.cpp
                src/models/fun_asr_nano/adaptor.cpp
        )
        if (AUDIOCPP_FUN_ASR_NANO_TEST_MODEL)
            add_test(
                NAME fun_asr_nano_adaptor_probe_cpu
                COMMAND fun_asr_nano_adaptor_probe
                    --backend cpu
                    --model ${AUDIOCPP_FUN_ASR_NANO_TEST_MODEL}
                    --reference ${CMAKE_CURRENT_SOURCE_DIR}/tests/fun_asr_nano/adaptor_reference.json
                    --data ${CMAKE_CURRENT_SOURCE_DIR}/tests/fun_asr_nano/adaptor_reference.bin
            )
            if (GGML_CUDA)
                add_test(
                    NAME fun_asr_nano_adaptor_probe_cuda
                    COMMAND fun_asr_nano_adaptor_probe
                        --backend cuda
                        --model ${AUDIOCPP_FUN_ASR_NANO_TEST_MODEL}
                        --reference ${CMAKE_CURRENT_SOURCE_DIR}/tests/fun_asr_nano/adaptor_reference.json
                        --data ${CMAKE_CURRENT_SOURCE_DIR}/tests/fun_asr_nano/adaptor_reference.bin
                )
            endif()
        endif()

        add_engine_unittest(fun_asr_nano_decoder_probe tests/fun_asr_nano/fun_asr_nano_decoder_probe.cpp)
        target_sources(
            fun_asr_nano_decoder_probe
            PRIVATE
                src/models/fun_asr_nano/assets.cpp
                src/models/fun_asr_nano/tokenizer_text.cpp
                src/models/fun_asr_nano/prompt.cpp
                src/models/fun_asr_nano/decoder.cpp
        )
        if (AUDIOCPP_FUN_ASR_NANO_TEST_MODEL)
            add_test(
                NAME fun_asr_nano_decoder_probe_cpu
                COMMAND fun_asr_nano_decoder_probe
                    --backend cpu
                    --model ${AUDIOCPP_FUN_ASR_NANO_TEST_MODEL}
                    --reference ${CMAKE_CURRENT_SOURCE_DIR}/tests/fun_asr_nano/decoder_reference.json
                    --data ${CMAKE_CURRENT_SOURCE_DIR}/tests/fun_asr_nano/decoder_reference.bin
            )
            if (GGML_CUDA)
                add_test(
                    NAME fun_asr_nano_decoder_probe_cuda
                    COMMAND fun_asr_nano_decoder_probe
                        --backend cuda
                        --model ${AUDIOCPP_FUN_ASR_NANO_TEST_MODEL}
                        --reference ${CMAKE_CURRENT_SOURCE_DIR}/tests/fun_asr_nano/decoder_reference.json
                        --data ${CMAKE_CURRENT_SOURCE_DIR}/tests/fun_asr_nano/decoder_reference.bin
                )
            endif()
        endif()

        add_engine_unittest(tdt_decoder_duration_loop_test tests/unittests/test_tdt_decoder_duration_loop.cpp)
        target_include_directories(tdt_decoder_duration_loop_test PRIVATE ${CMAKE_CURRENT_SOURCE_DIR}/tests/unittests)
        add_test(
            NAME tdt_decoder_duration_loop_test
            COMMAND tdt_decoder_duration_loop_test
        )

        if (hviske_asr IN_LIST AUDIOCPP_LINKED_MODELS)
        add_engine_unittest(asr_standalone_gguf_test tests/unittests/test_asr_standalone_gguf.cpp)
        target_sources(asr_standalone_gguf_test PRIVATE src/models/hviske_asr/assets.cpp)
        target_include_directories(asr_standalone_gguf_test PRIVATE ${CMAKE_CURRENT_SOURCE_DIR}/tests/unittests)
        target_compile_definitions(asr_standalone_gguf_test PRIVATE
            ENGINE_TEST_ASSET_ROOT="${ENGINE_UNITTEST_ASSET_ROOT}"
        )
        add_test(
            NAME asr_standalone_gguf_test
            COMMAND asr_standalone_gguf_test
        )
        endif()

        add_engine_unittest(parakeet_golden_transcription_test tests/parakeet_tdt/test_golden_transcription.cpp)
        target_compile_definitions(parakeet_golden_transcription_test PRIVATE
            ENGINE_REPO_ROOT="${CMAKE_CURRENT_SOURCE_DIR}"
        )
        add_test(
            NAME parakeet_golden_transcription_test
            COMMAND parakeet_golden_transcription_test
        )
        # Requires the real (multi-GB) Parakeet-TDT weights, which aren't
        # downloaded as part of a normal checkout/CI build; the test itself
        # detects this and exits with kExitSkip (125) rather than failing.
        set_tests_properties(parakeet_golden_transcription_test PROPERTIES
            SKIP_RETURN_CODE 125
            TIMEOUT 120
        )

        add_engine_unittest(parakeet_streaming_transcription_test tests/parakeet_tdt/test_streaming_transcription.cpp)
        target_compile_definitions(parakeet_streaming_transcription_test PRIVATE
            ENGINE_REPO_ROOT="${CMAKE_CURRENT_SOURCE_DIR}"
        )
        add_test(
            NAME parakeet_streaming_transcription_test
            COMMAND parakeet_streaming_transcription_test
        )
        set_tests_properties(parakeet_streaming_transcription_test PROPERTIES
            SKIP_RETURN_CODE 125
            TIMEOUT 180
        )

        # Numerical parity dump tool (not registered as an add_test -- it's a
        # manual/pre-release check, not a CI gate). Run it against a NeMo
        # reference dump via compare_parity.py; see tests/parakeet_tdt/parity/README.md.
        if (parakeet_tdt IN_LIST AUDIOCPP_LINKED_MODELS)
        add_executable(parakeet_parity_dump
            tests/parakeet_tdt/parity/dump_cpp_reference.cpp
        )
        target_include_directories(parakeet_parity_dump PRIVATE ${CMAKE_CURRENT_SOURCE_DIR}/tests/parakeet_tdt/parity)
        target_link_libraries(parakeet_parity_dump PRIVATE engine_runtime ggml)
        if (ENGINE_ENABLE_OPENMP)
            target_link_libraries(parakeet_parity_dump PRIVATE OpenMP::OpenMP_CXX)
        endif()
        endif()
    endif()
endif()

if (ENGINE_BUILD_EXTENDED_TESTS AND ENGINE_ENABLE_CUDA)
    add_executable(cuda_conv_transpose_1d_test tests/unittests/test_cuda_conv_transpose_1d.cu)
    set_target_properties(cuda_conv_transpose_1d_test PROPERTIES CUDA_STANDARD 17 CUDA_STANDARD_REQUIRED ON)
    target_compile_options(cuda_conv_transpose_1d_test PRIVATE $<$<COMPILE_LANGUAGE:CUDA>:-use_fast_math>)
    if (GGML_STATIC)
        set_target_properties(cuda_conv_transpose_1d_test PROPERTIES CUDA_RUNTIME_LIBRARY Static)
        target_link_libraries(cuda_conv_transpose_1d_test PRIVATE ggml CUDA::cudart_static)
    else()
        set_target_properties(cuda_conv_transpose_1d_test PROPERTIES CUDA_RUNTIME_LIBRARY Shared)
        target_link_libraries(cuda_conv_transpose_1d_test PRIVATE ggml CUDA::cudart)
    endif()
    add_test(NAME cuda_conv_transpose_1d_test COMMAND cuda_conv_transpose_1d_test)
    set_tests_properties(cuda_conv_transpose_1d_test PROPERTIES SKIP_RETURN_CODE 77 TIMEOUT 120)
endif()

if (ENGINE_BUILD_EXAMPLES)
    add_executable(ggml_simple_inference
        examples/ggml_simple_inference.cpp
    )

    target_link_libraries(ggml_simple_inference PRIVATE ggml)

    add_executable(model_spec_demo
        examples/model_spec_demo/model_spec_demo.cpp
    )
    target_link_libraries(model_spec_demo PRIVATE engine_runtime ggml)
    if (ENGINE_ENABLE_OPENMP)
        target_link_libraries(model_spec_demo PRIVATE OpenMP::OpenMP_CXX)
    endif()

    add_executable(model_spec_download_demo
        examples/model_spec_demo/model_spec_download_demo.cpp
    )
    target_link_libraries(model_spec_download_demo PRIVATE engine_runtime ggml)
    if (ENGINE_ENABLE_OPENMP)
        target_link_libraries(model_spec_download_demo PRIVATE OpenMP::OpenMP_CXX)
    endif()
endif()