feat(backend): add moss-tts-cpp text-to-speech backend (#10860)

* feat(backend): add moss-tts-cpp text-to-speech backend Add a Go + purego backend wrapping the moss-tts.cpp ggml port of the OpenMOSS MOSS-TTS-Local v1.5 text-to-speech model (GPT-J local transformer decoded through MOSS-Audio-Tokenizer-v2), producing 48 kHz stereo audio with optional reference-audio voice cloning. Mirrors the qwen3-tts-cpp backend: dlopen the static-ggml shared library, bind the moss-tts.cpp C-API via purego, and serve the gRPC TTS method. A thin C shim holds the pipeline handle and copies engine PCM into a Go-freeable buffer. Wires the CI registration: backend-matrix.yml (CPU, CUDA 12/13, Intel SYCL f16/f32, Vulkan, ROCm, NVIDIA L4T, plus Darwin metal), backend/index.yaml metas and image entries pointing at mudler/MOSS-TTS-Local-Transformer-v1.5-GGUF, the root Makefile build targets, and the changed-backends.js path mapping. Assisted-by: Claude:claude-opus-4-8 [Claude Code] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * docs: list the moss-tts-cpp backend among the LocalAI-maintained engines Add moss-tts.cpp to the README "Backends built by us" table, the Text-to-Speech compatibility table, and the reference-audio voice-cloning backend list, so the new backend is documented alongside its peers. Assisted-by: Claude:claude-opus-4-8 [Claude Code] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * backend(moss-tts-cpp): pin moss-tts.cpp to the squashed single-commit release moss-tts.cpp history was collapsed to a single commit; repoint MOSSTTS_CPP_VERSION to ee722b8e9205ee9b1b1c398a4e87e4e393e9be41. Assisted-by: Claude:claude-opus-4-8 [Claude Code] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * backend(moss-tts-cpp): add the moss-tts-cpp-development gallery meta The gallery had the -development image entries but no matching -development meta anchor (as locate-anything-cpp and depth-anything-cpp have), so the master build was not installable as a gallery backend. Add moss-tts-cpp-development mirroring the production meta with the -development capability image names. Assisted-by: Claude:claude-opus-4-8 [Claude Code] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> --------- Signed-off-by: Ettore Di Giacinto <mudler@localai.io> Co-authored-by: Ettore Di Giacinto <mudler@localai.io>
2026-07-30 09:57:57 -04:00 · 2026-07-17 09:26:12 +02:00
parent 0bd7a29f31
commit 3bb0d1cb49
23 changed files with 1591 additions and 3 deletions
--- a/.github/backend-matrix.yml
+++ b/.github/backend-matrix.yml
@@ -858,6 +858,19 @@ include:
    dockerfile: "./backend/Dockerfile.golang"
    context: "./"
    ubuntu-version: '2404'
+  - build-type: 'cublas'
+    cuda-major-version: "12"
+    cuda-minor-version: "8"
+    platforms: 'linux/amd64'
+    tag-latest: 'auto'
+    tag-suffix: '-gpu-nvidia-cuda-12-moss-tts-cpp'
+    runs-on: 'ubuntu-latest'
+    base-image: "ubuntu:24.04"
+    skip-drivers: 'false'
+    backend: "moss-tts-cpp"
+    dockerfile: "./backend/Dockerfile.golang"
+    context: "./"
+    ubuntu-version: '2404'
  - build-type: 'cublas'
    cuda-major-version: "12"
    cuda-minor-version: "8"
@@ -1909,6 +1922,19 @@ include:
    dockerfile: "./backend/Dockerfile.golang"
    context: "./"
    ubuntu-version: '2404'
+  - build-type: 'cublas'
+    cuda-major-version: "13"
+    cuda-minor-version: "0"
+    platforms: 'linux/amd64'
+    tag-latest: 'auto'
+    tag-suffix: '-gpu-nvidia-cuda-13-moss-tts-cpp'
+    runs-on: 'ubuntu-latest'
+    base-image: "ubuntu:24.04"
+    skip-drivers: 'false'
+    backend: "moss-tts-cpp"
+    dockerfile: "./backend/Dockerfile.golang"
+    context: "./"
+    ubuntu-version: '2404'
  - build-type: 'cublas'
    cuda-major-version: "13"
    cuda-minor-version: "0"
@@ -1961,6 +1987,19 @@ include:
    backend: "qwen3-tts-cpp"
    dockerfile: "./backend/Dockerfile.golang"
    context: "./"
+  - build-type: 'cublas'
+    cuda-major-version: "13"
+    cuda-minor-version: "0"
+    platforms: 'linux/arm64'
+    skip-drivers: 'false'
+    tag-latest: 'auto'
+    tag-suffix: '-nvidia-l4t-cuda-13-arm64-moss-tts-cpp'
+    base-image: "ubuntu:24.04"
+    ubuntu-version: '2404'
+    runs-on: 'ubuntu-24.04-arm'
+    backend: "moss-tts-cpp"
+    dockerfile: "./backend/Dockerfile.golang"
+    context: "./"
  - build-type: 'cublas'
    cuda-major-version: "13"
    cuda-minor-version: "0"
@@ -4529,6 +4568,35 @@ include:
    dockerfile: "./backend/Dockerfile.golang"
    context: "./"
    ubuntu-version: '2404'
+  # moss-tts-cpp
+  - build-type: ''
+    cuda-major-version: ""
+    cuda-minor-version: ""
+    platforms: 'linux/amd64'
+    platform-tag: 'amd64'
+    tag-latest: 'auto'
+    tag-suffix: '-cpu-moss-tts-cpp'
+    runs-on: 'ubuntu-latest'
+    base-image: "ubuntu:24.04"
+    skip-drivers: 'false'
+    backend: "moss-tts-cpp"
+    dockerfile: "./backend/Dockerfile.golang"
+    context: "./"
+    ubuntu-version: '2404'
+  - build-type: ''
+    cuda-major-version: ""
+    cuda-minor-version: ""
+    platforms: 'linux/arm64'
+    platform-tag: 'arm64'
+    tag-latest: 'auto'
+    tag-suffix: '-cpu-moss-tts-cpp'
+    runs-on: 'ubuntu-24.04-arm'
+    base-image: "ubuntu:24.04"
+    skip-drivers: 'false'
+    backend: "moss-tts-cpp"
+    dockerfile: "./backend/Dockerfile.golang"
+    context: "./"
+    ubuntu-version: '2404'
  # omnivoice-cpp
  - build-type: ''
    cuda-major-version: ""
@@ -4571,6 +4639,19 @@ include:
    dockerfile: "./backend/Dockerfile.golang"
    context: "./"
    ubuntu-version: '2404'
+  - build-type: 'sycl_f32'
+    cuda-major-version: ""
+    cuda-minor-version: ""
+    platforms: 'linux/amd64'
+    tag-latest: 'auto'
+    tag-suffix: '-gpu-intel-sycl-f32-moss-tts-cpp'
+    runs-on: 'ubuntu-latest'
+    base-image: "intel/oneapi-basekit:2025.3.0-0-devel-ubuntu24.04"
+    skip-drivers: 'false'
+    backend: "moss-tts-cpp"
+    dockerfile: "./backend/Dockerfile.golang"
+    context: "./"
+    ubuntu-version: '2404'
  - build-type: 'sycl_f32'
    cuda-major-version: ""
    cuda-minor-version: ""
@@ -4597,6 +4678,19 @@ include:
    dockerfile: "./backend/Dockerfile.golang"
    context: "./"
    ubuntu-version: '2404'
+  - build-type: 'sycl_f16'
+    cuda-major-version: ""
+    cuda-minor-version: ""
+    platforms: 'linux/amd64'
+    tag-latest: 'auto'
+    tag-suffix: '-gpu-intel-sycl-f16-moss-tts-cpp'
+    runs-on: 'ubuntu-latest'
+    base-image: "intel/oneapi-basekit:2025.3.0-0-devel-ubuntu24.04"
+    skip-drivers: 'false'
+    backend: "moss-tts-cpp"
+    dockerfile: "./backend/Dockerfile.golang"
+    context: "./"
+    ubuntu-version: '2404'
  - build-type: 'sycl_f16'
    cuda-major-version: ""
    cuda-minor-version: ""
@@ -4624,6 +4718,20 @@ include:
    dockerfile: "./backend/Dockerfile.golang"
    context: "./"
    ubuntu-version: '2404'
+  - build-type: 'vulkan'
+    cuda-major-version: ""
+    cuda-minor-version: ""
+    platforms: 'linux/amd64'
+    platform-tag: 'amd64'
+    tag-latest: 'auto'
+    tag-suffix: '-gpu-vulkan-moss-tts-cpp'
+    runs-on: 'ubuntu-latest'
+    base-image: "ubuntu:24.04"
+    skip-drivers: 'false'
+    backend: "moss-tts-cpp"
+    dockerfile: "./backend/Dockerfile.golang"
+    context: "./"
+    ubuntu-version: '2404'
  - build-type: 'vulkan'
    cuda-major-version: ""
    cuda-minor-version: ""
@@ -4652,6 +4760,20 @@ include:
    dockerfile: "./backend/Dockerfile.golang"
    context: "./"
    ubuntu-version: '2404'
+  - build-type: 'vulkan'
+    cuda-major-version: ""
+    cuda-minor-version: ""
+    platforms: 'linux/arm64'
+    platform-tag: 'arm64'
+    tag-latest: 'auto'
+    tag-suffix: '-gpu-vulkan-moss-tts-cpp'
+    runs-on: 'ubuntu-24.04-arm'
+    base-image: "ubuntu:24.04"
+    skip-drivers: 'false'
+    backend: "moss-tts-cpp"
+    dockerfile: "./backend/Dockerfile.golang"
+    context: "./"
+    ubuntu-version: '2404'
  - build-type: 'vulkan'
    cuda-major-version: ""
    cuda-minor-version: ""
@@ -4679,6 +4801,19 @@ include:
    dockerfile: "./backend/Dockerfile.golang"
    context: "./"
    ubuntu-version: '2204'
+  - build-type: 'cublas'
+    cuda-major-version: "12"
+    cuda-minor-version: "0"
+    platforms: 'linux/arm64'
+    skip-drivers: 'false'
+    tag-latest: 'auto'
+    tag-suffix: '-nvidia-l4t-arm64-moss-tts-cpp'
+    base-image: "nvcr.io/nvidia/l4t-jetpack:r36.4.0"
+    runs-on: 'ubuntu-24.04-arm'
+    backend: "moss-tts-cpp"
+    dockerfile: "./backend/Dockerfile.golang"
+    context: "./"
+    ubuntu-version: '2204'
  - build-type: 'cublas'
    cuda-major-version: "12"
    cuda-minor-version: "0"
@@ -4705,6 +4840,19 @@ include:
    dockerfile: "./backend/Dockerfile.golang"
    context: "./"
    ubuntu-version: '2404'
+  - build-type: 'hipblas'
+    cuda-major-version: ""
+    cuda-minor-version: ""
+    platforms: 'linux/amd64'
+    tag-latest: 'auto'
+    tag-suffix: '-gpu-rocm-hipblas-moss-tts-cpp'
+    base-image: "rocm/dev-ubuntu-24.04:6.4.4"
+    runs-on: 'ubuntu-latest'
+    skip-drivers: 'false'
+    backend: "moss-tts-cpp"
+    dockerfile: "./backend/Dockerfile.golang"
+    context: "./"
+    ubuntu-version: '2404'
  - build-type: 'hipblas'
    cuda-major-version: ""
    cuda-minor-version: ""
@@ -5622,6 +5770,10 @@ includeDarwin:
    tag-suffix: "-metal-darwin-arm64-qwen3-tts-cpp"
    build-type: "metal"
    lang: "go"
+  - backend: "moss-tts-cpp"
+    tag-suffix: "-metal-darwin-arm64-moss-tts-cpp"
+    build-type: "metal"
+    lang: "go"
  - backend: "omnivoice-cpp"
    tag-suffix: "-metal-darwin-arm64-omnivoice-cpp"
    build-type: "metal"
--- a/6
+++ b/6
@@ -1,5 +1,5 @@
 # Disable parallel execution for backend builds
-.NOTPARALLEL: backends/diffusers backends/llama-cpp backends/turboquant backends/bonsai backends/outetts backends/piper backends/stablediffusion-ggml backends/whisper backends/crispasr backends/parakeet-cpp backends/moss-transcribe-cpp backends/faster-whisper backends/silero-vad backends/local-store backends/cloud-proxy backends/huggingface backends/rfdetr backends/rfdetr-cpp backends/insightface backends/speaker-recognition backends/kitten-tts backends/kokoro backends/chatterbox backends/llama-cpp-darwin backends/neutts build-darwin-python-backend build-darwin-go-backend backends/mlx backends/diffuser-darwin backends/mlx-vlm backends/mlx-audio backends/mlx-distributed backends/stablediffusion-ggml-darwin backends/vllm backends/vllm-omni backends/longcat-video backends/sglang backends/moonshine backends/pocket-tts backends/qwen-tts backends/faster-qwen3-tts backends/qwen-asr backends/nemo backends/voxcpm backends/whisperx backends/ace-step backends/acestep-cpp backends/fish-speech backends/voxtral backends/opus backends/trl backends/llama-cpp-quantization backends/kokoros backends/sam3-cpp backends/qwen3-tts-cpp backends/omnivoice-cpp backends/vibevoice-cpp backends/localvqe backends/tinygrad backends/sherpa-onnx backends/ds4 backends/ds4-darwin backends/liquid-audio backends/supertonic backends/depth-anything-cpp backends/privacy-filter backends/privacy-filter-darwin
+.NOTPARALLEL: backends/diffusers backends/llama-cpp backends/turboquant backends/bonsai backends/outetts backends/piper backends/stablediffusion-ggml backends/whisper backends/crispasr backends/parakeet-cpp backends/moss-transcribe-cpp backends/faster-whisper backends/silero-vad backends/local-store backends/cloud-proxy backends/huggingface backends/rfdetr backends/rfdetr-cpp backends/insightface backends/speaker-recognition backends/kitten-tts backends/kokoro backends/chatterbox backends/llama-cpp-darwin backends/neutts build-darwin-python-backend build-darwin-go-backend backends/mlx backends/diffuser-darwin backends/mlx-vlm backends/mlx-audio backends/mlx-distributed backends/stablediffusion-ggml-darwin backends/vllm backends/vllm-omni backends/longcat-video backends/sglang backends/moonshine backends/pocket-tts backends/qwen-tts backends/faster-qwen3-tts backends/qwen-asr backends/nemo backends/voxcpm backends/whisperx backends/ace-step backends/acestep-cpp backends/fish-speech backends/voxtral backends/opus backends/trl backends/llama-cpp-quantization backends/kokoros backends/sam3-cpp backends/qwen3-tts-cpp backends/moss-tts-cpp backends/omnivoice-cpp backends/vibevoice-cpp backends/localvqe backends/tinygrad backends/sherpa-onnx backends/ds4 backends/ds4-darwin backends/liquid-audio backends/supertonic backends/depth-anything-cpp backends/privacy-filter backends/privacy-filter-darwin

 GOCMD=go
 GOTEST=$(GOCMD) test
@@ -1249,6 +1249,7 @@ BACKEND_DEPTH_ANYTHING_CPP = depth-anything-cpp|golang|.|false|true
 BACKEND_VOXTRAL = voxtral|golang|.|false|true
 BACKEND_ACESTEP_CPP = acestep-cpp|golang|.|false|true
 BACKEND_QWEN3_TTS_CPP = qwen3-tts-cpp|golang|.|false|true
+BACKEND_MOSS_TTS_CPP = moss-tts-cpp|golang|.|false|true
 BACKEND_OMNIVOICE_CPP = omnivoice-cpp|golang|.|false|true
 BACKEND_VIBEVOICE_CPP = vibevoice-cpp|golang|.|false|true
 BACKEND_LOCALVQE = localvqe|golang|.|false|true
@@ -1375,6 +1376,7 @@ $(eval $(call generate-docker-build-target,$(BACKEND_WHISPERX)))
 $(eval $(call generate-docker-build-target,$(BACKEND_ACE_STEP)))
 $(eval $(call generate-docker-build-target,$(BACKEND_ACESTEP_CPP)))
 $(eval $(call generate-docker-build-target,$(BACKEND_QWEN3_TTS_CPP)))
+$(eval $(call generate-docker-build-target,$(BACKEND_MOSS_TTS_CPP)))
 $(eval $(call generate-docker-build-target,$(BACKEND_OMNIVOICE_CPP)))
 $(eval $(call generate-docker-build-target,$(BACKEND_VIBEVOICE_CPP)))
 $(eval $(call generate-docker-build-target,$(BACKEND_LOCALVQE)))
@@ -1394,7 +1396,7 @@ $(eval $(call generate-docker-build-target,$(BACKEND_SUPERTONIC)))
 docker-save-%: backend-images
 	docker save local-ai-backend:$* -o backend-images/$*.tar

-docker-build-backends: docker-build-llama-cpp docker-build-ik-llama-cpp docker-build-turboquant docker-build-bonsai docker-build-ds4 docker-build-rerankers docker-build-vllm docker-build-vllm-omni docker-build-longcat-video docker-build-sglang docker-build-transformers docker-build-outetts docker-build-diffusers docker-build-kokoro docker-build-faster-whisper docker-build-crispasr docker-build-coqui docker-build-chatterbox docker-build-vibevoice docker-build-liquid-audio docker-build-moonshine docker-build-pocket-tts docker-build-qwen-tts docker-build-fish-speech docker-build-faster-qwen3-tts docker-build-qwen-asr docker-build-nemo docker-build-voxcpm docker-build-whisperx docker-build-ace-step docker-build-acestep-cpp docker-build-voxtral docker-build-mlx-distributed docker-build-trl docker-build-llama-cpp-quantization docker-build-tinygrad docker-build-kokoros docker-build-sam3-cpp docker-build-rfdetr-cpp docker-build-qwen3-tts-cpp docker-build-omnivoice-cpp docker-build-vibevoice-cpp docker-build-localvqe docker-build-insightface docker-build-speaker-recognition docker-build-sherpa-onnx docker-build-cloud-proxy docker-build-supertonic docker-build-depth-anything-cpp docker-build-moss-transcribe-cpp docker-build-privacy-filter
+docker-build-backends: docker-build-llama-cpp docker-build-ik-llama-cpp docker-build-turboquant docker-build-bonsai docker-build-ds4 docker-build-rerankers docker-build-vllm docker-build-vllm-omni docker-build-longcat-video docker-build-sglang docker-build-transformers docker-build-outetts docker-build-diffusers docker-build-kokoro docker-build-faster-whisper docker-build-crispasr docker-build-coqui docker-build-chatterbox docker-build-vibevoice docker-build-liquid-audio docker-build-moonshine docker-build-pocket-tts docker-build-qwen-tts docker-build-fish-speech docker-build-faster-qwen3-tts docker-build-qwen-asr docker-build-nemo docker-build-voxcpm docker-build-whisperx docker-build-ace-step docker-build-acestep-cpp docker-build-voxtral docker-build-mlx-distributed docker-build-trl docker-build-llama-cpp-quantization docker-build-tinygrad docker-build-kokoros docker-build-sam3-cpp docker-build-rfdetr-cpp docker-build-qwen3-tts-cpp docker-build-moss-tts-cpp docker-build-omnivoice-cpp docker-build-vibevoice-cpp docker-build-localvqe docker-build-insightface docker-build-speaker-recognition docker-build-sherpa-onnx docker-build-cloud-proxy docker-build-supertonic docker-build-depth-anything-cpp docker-build-moss-transcribe-cpp docker-build-privacy-filter

 ########################################################
 ### Mock Backend for E2E Tests
--- a/README.md
+++ b/README.md
@@ -233,6 +233,7 @@ Most backends wrap a best-in-class upstream engine. A handful of them are native
 |---------|-------------|
 | [parakeet.cpp](https://github.com/mudler/parakeet.cpp) | C++/GGML port of NVIDIA NeMo Parakeet ASR (tdt/ctc/rnnt/hybrid), with cache-aware streaming transcription |
 | [moss-transcribe.cpp](https://github.com/mudler/moss-transcribe.cpp) | C++/GGML port of OpenMOSS MOSS-Transcribe-Diarize: joint long-form transcription, speaker diarization and timestamping in a single pass |
+| [moss-tts.cpp](https://github.com/mudler/moss-tts.cpp) | C++/GGML port of the OpenMOSS MOSS-TTS family: text-to-speech (MOSS-TTS-Local v1.5, 48 kHz stereo) with reference-audio voice cloning, through the MOSS-Audio-Tokenizer neural codec |
 | [ced.cpp](https://github.com/mudler/ced.cpp) | C++/GGML port of the CED audio-tagging models: sound-event classification (527-class AudioSet) over REST and the realtime API for live recognition |
 | [voice-detect.cpp](https://github.com/mudler/voice-detect.cpp) | Speaker recognition and voice analysis (ECAPA-TDNN, WeSpeaker, ERes2Net, CAM++, wav2vec2 age/gender/emotion), replacing the Python speaker-recognition backend |
 | [voxtral-tts.c](https://github.com/mudler/voxtral-tts.c) | Voxtral Realtime 4B speech-to-text in pure C |
--- a/backend/go/moss-tts-cpp/.gitignore
+++ b/backend/go/moss-tts-cpp/.gitignore
@@ -0,0 +1,5 @@
+moss-tts-cpp
+*.so
+*.dylib
+sources/
+package/
--- a/backend/go/moss-tts-cpp/CMakeLists.txt
+++ b/backend/go/moss-tts-cpp/CMakeLists.txt
@@ -0,0 +1,48 @@
+cmake_minimum_required(VERSION 3.18)
+project(gomosstts-cpp LANGUAGES C CXX)
+set(CMAKE_POSITION_INDEPENDENT_CODE ON)
+set(CMAKE_EXPORT_COMPILE_COMMANDS ON)
+
+set(MOSS_DIR ${CMAKE_CURRENT_SOURCE_DIR}/sources/moss-tts.cpp)
+
+# Override upstream's CMAKE_CUDA_ARCHITECTURES before add_subdirectory.
+if(NOT DEFINED CMAKE_CUDA_ARCHITECTURES)
+    set(CMAKE_CUDA_ARCHITECTURES "75-virtual;80-virtual;86-real;89-real")
+endif()
+
+# Add the upstream project. Its own CMakeLists vendors ggml under
+# third_party/ggml and builds the moss-tts library (STATIC by default, since we
+# do NOT set MOSS_TTS_SHARED). EXCLUDE_FROM_ALL keeps its CLI tool / tests /
+# examples from building unless referenced.
+add_subdirectory(${MOSS_DIR} mosstts EXCLUDE_FROM_ALL)
+
+add_library(gomosstts-cpp MODULE cpp/mossttscpp.cpp)
+target_link_libraries(gomosstts-cpp PRIVATE moss-tts)
+
+target_include_directories(gomosstts-cpp PRIVATE ${MOSS_DIR}/include ${MOSS_DIR}/src)
+target_include_directories(gomosstts-cpp SYSTEM PRIVATE ${MOSS_DIR}/third_party/ggml/include)
+
+# Link GPU backends if the upstream ggml created them. moss-tts links these
+# PUBLIC when enabled, but we re-list them so the MODULE carries the symbols
+# even under static ggml (mirrors qwen3-tts-cpp).
+foreach(backend blas cuda hip metal vulkan sycl)
+    if(TARGET ggml-${backend})
+        target_link_libraries(gomosstts-cpp PRIVATE ggml-${backend})
+        if(backend STREQUAL "cuda")
+            find_package(CUDAToolkit QUIET)
+            if(CUDAToolkit_FOUND)
+                target_link_libraries(gomosstts-cpp PRIVATE CUDA::cudart)
+            endif()
+        endif()
+    endif()
+endforeach()
+
+if(MSVC)
+    target_compile_options(gomosstts-cpp PRIVATE /W4 /wd4100 /wd4505)
+else()
+    target_compile_options(gomosstts-cpp PRIVATE -Wall -Wextra
+                          -Wno-unused-parameter -Wno-unused-function)
+endif()
+
+set_property(TARGET gomosstts-cpp PROPERTY CXX_STANDARD 17)
+set_target_properties(gomosstts-cpp PROPERTIES LIBRARY_OUTPUT_DIRECTORY ${CMAKE_BINARY_DIR})
--- a/backend/go/moss-tts-cpp/Makefile
+++ b/backend/go/moss-tts-cpp/Makefile
@@ -0,0 +1,140 @@
+CMAKE_ARGS?=
+BUILD_TYPE?=
+NATIVE?=false
+
+GOCMD?=go
+GO_TAGS?=
+JOBS?=$(shell nproc --ignore=1)
+
+# moss-tts.cpp version
+MOSSTTS_REPO?=https://github.com/mudler/moss-tts.cpp
+MOSSTTS_CPP_VERSION?=ee722b8e9205ee9b1b1c398a4e87e4e393e9be41
+SO_TARGET?=libgomosstts-cpp.so
+
+CMAKE_ARGS+=-DBUILD_SHARED_LIBS=OFF
+
+ifeq ($(NATIVE),false)
+	CMAKE_ARGS+=-DGGML_NATIVE=OFF
+endif
+
+ifeq ($(BUILD_TYPE),cublas)
+	CMAKE_ARGS+=-DGGML_CUDA=ON
+else ifeq ($(BUILD_TYPE),openblas)
+	CMAKE_ARGS+=-DGGML_BLAS=ON -DGGML_BLAS_VENDOR=OpenBLAS
+else ifeq ($(BUILD_TYPE),clblas)
+	CMAKE_ARGS+=-DGGML_CLBLAST=ON -DCLBlast_DIR=/some/path
+else ifeq ($(BUILD_TYPE),hipblas)
+	# This ggml only understands GGML_HIP (GGML_HIPBLAS was removed upstream),
+	# so passing GGML_HIPBLAS silently produced a CPU-only build (see #10666).
+	ROCM_HOME ?= /opt/rocm
+	ROCM_PATH ?= /opt/rocm
+	export CXX=$(ROCM_HOME)/llvm/bin/clang++
+	export CC=$(ROCM_HOME)/llvm/bin/clang
+	AMDGPU_TARGETS ?= gfx908,gfx90a,gfx942,gfx950,gfx1030,gfx1100,gfx1101,gfx1102,gfx1151,gfx1200,gfx1201
+	CMAKE_ARGS+=-DGGML_HIP=ON -DAMDGPU_TARGETS=$(AMDGPU_TARGETS)
+else ifeq ($(BUILD_TYPE),vulkan)
+	CMAKE_ARGS+=-DGGML_VULKAN=ON
+else ifeq ($(OS),Darwin)
+	ifneq ($(BUILD_TYPE),metal)
+		CMAKE_ARGS+=-DGGML_METAL=OFF
+	else
+		CMAKE_ARGS+=-DGGML_METAL=ON
+		CMAKE_ARGS+=-DGGML_METAL_EMBED_LIBRARY=ON
+	endif
+endif
+
+ifeq ($(BUILD_TYPE),sycl_f16)
+	CMAKE_ARGS+=-DGGML_SYCL=ON \
+		-DCMAKE_C_COMPILER=icx \
+		-DCMAKE_CXX_COMPILER=icpx \
+		-DGGML_SYCL_F16=ON
+endif
+
+ifeq ($(BUILD_TYPE),sycl_f32)
+	CMAKE_ARGS+=-DGGML_SYCL=ON \
+		-DCMAKE_C_COMPILER=icx \
+		-DCMAKE_CXX_COMPILER=icpx
+endif
+
+sources/moss-tts.cpp:
+	mkdir -p sources/moss-tts.cpp
+	cd sources/moss-tts.cpp && \
+	git init && \
+	git remote add origin $(MOSSTTS_REPO) && \
+	git fetch origin && \
+	git checkout $(MOSSTTS_CPP_VERSION) && \
+	git submodule update --init --recursive --depth 1 --single-branch
+
+# Detect OS
+UNAME_S := $(shell uname -s)
+
+# Only build CPU variants on Linux
+ifeq ($(UNAME_S),Linux)
+	VARIANT_TARGETS = libgomosstts-cpp-avx.so libgomosstts-cpp-avx2.so libgomosstts-cpp-avx512.so libgomosstts-cpp-fallback.so
+else
+	# On non-Linux (e.g., Darwin), build only fallback variant (as a dylib)
+	VARIANT_TARGETS = libgomosstts-cpp-fallback.dylib
+endif
+
+moss-tts-cpp: main.go gomossttscpp.go $(VARIANT_TARGETS)
+	CGO_ENABLED=0 $(GOCMD) build -tags "$(GO_TAGS)" -o moss-tts-cpp ./
+
+package: moss-tts-cpp
+	bash package.sh
+
+build: package
+
+clean: purge
+	rm -rf libgomosstts-cpp*.so libgomosstts-cpp*.dylib package sources/moss-tts.cpp moss-tts-cpp
+
+purge:
+	rm -rf build*
+
+# Variants must build sequentially
+.NOTPARALLEL:
+
+# Build all variants (Linux only)
+ifeq ($(UNAME_S),Linux)
+libgomosstts-cpp-avx.so: sources/moss-tts.cpp
+	$(info ${GREEN}I moss-tts-cpp build info:avx${RESET})
+	SO_TARGET=libgomosstts-cpp-avx.so CMAKE_ARGS="$(CMAKE_ARGS) -DGGML_AVX=on -DGGML_AVX2=off -DGGML_AVX512=off -DGGML_FMA=off -DGGML_F16C=off -DGGML_BMI2=off" $(MAKE) libgomosstts-cpp-custom
+	rm -rf build-libgomosstts-cpp-avx.so
+
+libgomosstts-cpp-avx2.so: sources/moss-tts.cpp
+	$(info ${GREEN}I moss-tts-cpp build info:avx2${RESET})
+	SO_TARGET=libgomosstts-cpp-avx2.so CMAKE_ARGS="$(CMAKE_ARGS) -DGGML_AVX=on -DGGML_AVX2=on -DGGML_AVX512=off -DGGML_FMA=on -DGGML_F16C=on -DGGML_BMI2=on" $(MAKE) libgomosstts-cpp-custom
+	rm -rf build-libgomosstts-cpp-avx2.so
+
+libgomosstts-cpp-avx512.so: sources/moss-tts.cpp
+	$(info ${GREEN}I moss-tts-cpp build info:avx512${RESET})
+	SO_TARGET=libgomosstts-cpp-avx512.so CMAKE_ARGS="$(CMAKE_ARGS) -DGGML_AVX=on -DGGML_AVX2=on -DGGML_AVX512=on -DGGML_FMA=on -DGGML_F16C=on -DGGML_BMI2=on" $(MAKE) libgomosstts-cpp-custom
+	rm -rf build-libgomosstts-cpp-avx512.so
+endif
+
+# Build fallback variant (all platforms)
+libgomosstts-cpp-fallback.so: sources/moss-tts.cpp
+	$(info ${GREEN}I moss-tts-cpp build info:fallback${RESET})
+	SO_TARGET=libgomosstts-cpp-fallback.so CMAKE_ARGS="$(CMAKE_ARGS) -DGGML_AVX=off -DGGML_AVX2=off -DGGML_AVX512=off -DGGML_FMA=off -DGGML_F16C=off -DGGML_BMI2=off" $(MAKE) libgomosstts-cpp-custom
+	rm -rf build-libgomosstts-cpp-fallback.so
+
+# Build fallback variant as a dylib (Darwin)
+libgomosstts-cpp-fallback.dylib: sources/moss-tts.cpp
+	$(info ${GREEN}I moss-tts-cpp build info:fallback (dylib)${RESET})
+	SO_TARGET=libgomosstts-cpp-fallback.dylib CMAKE_ARGS="$(CMAKE_ARGS) -DGGML_AVX=off -DGGML_AVX2=off -DGGML_AVX512=off -DGGML_FMA=off -DGGML_F16C=off -DGGML_BMI2=off" $(MAKE) libgomosstts-cpp-custom
+	rm -rf build-libgomosstts-cpp-fallback.dylib
+
+libgomosstts-cpp-custom: CMakeLists.txt cpp/mossttscpp.cpp cpp/mossttscpp.h
+	mkdir -p build-$(SO_TARGET) && \
+	cd build-$(SO_TARGET) && \
+	cmake .. $(CMAKE_ARGS) && \
+	cmake --build . --config Release -j$(JOBS) --target gomosstts-cpp && \
+	cd .. && \
+	(mv build-$(SO_TARGET)/libgomosstts-cpp.so ./$(SO_TARGET) 2>/dev/null || \
+	 mv build-$(SO_TARGET)/libgomosstts-cpp.dylib ./$(SO_TARGET) 2>/dev/null)
+
+test: moss-tts-cpp
+	@echo "Running moss-tts-cpp tests..."
+	bash test.sh
+	@echo "moss-tts-cpp tests completed."
+
+all: moss-tts-cpp package
--- a/backend/go/moss-tts-cpp/README.md
+++ b/backend/go/moss-tts-cpp/README.md
@@ -0,0 +1,79 @@
+# MOSS-TTS C++ backend
+
+This backend runs the OpenMOSS **MOSS-TTS-Local (v1.5)** GGUF model through
+[moss-tts.cpp](https://github.com/mudler/moss-tts.cpp), a from-scratch C++/ggml
+port with no Python at inference time. It generates **48 kHz stereo** speech and
+supports reference-audio voice cloning.
+
+The engine loads three GGUFs: the local transformer (the model), the
+MOSS-Audio-Tokenizer neural codec, and the text tokenizer. It is loaded via
+purego (cgo-less `dlopen`) exactly like `qwen3-tts-cpp`.
+
+## Model configuration
+
+The model path points at the local transformer GGUF. The codec and text
+tokenizer are auto-discovered as siblings of the model:
+
+- codec: a `*.gguf` whose name contains `audio` + `tokenizer` (or `codec`),
+  e.g. `moss-audio-tokenizer-v2-f32.gguf`
+- tokenizer: the other `*.gguf` whose name contains `tokenizer`,
+  e.g. `moss-tokenizer-v1_5.gguf`
+
+```yaml
+name: moss-tts-cpp
+backend: moss-tts-cpp
+parameters:
+  model: moss-tts-local-v1_5-q8_0.gguf
+known_usecases:
+  - tts
+tts:
+  audio_path: voices/default-reference.wav  # optional model-wide clone reference
+```
+
+Override discovery when the filenames are non-standard:
+
+```yaml
+options:
+  - "codec:moss-audio-tokenizer-v2-f32.gguf"
+  - "tokenizer:moss-tokenizer-v1_5.gguf"
+  - "seed:42"
+```
+
+GGUFs for the Local v1.5 model (plus its codec and tokenizer) live at
+[mudler/MOSS-TTS-Local-Transformer-v1.5-GGUF](https://huggingface.co/mudler/MOSS-TTS-Local-Transformer-v1.5-GGUF).
+
+## Voice cloning
+
+MOSS-TTS-Local has no named speakers; cloning is driven purely by a
+reference-audio **path**. Request precedence is: a request `voice` that ends in
+a known audio extension (`.wav`, `.flac`, `.mp3`, `.ogg`, `.m4a`), then
+`tts.audio_path`. The engine decodes the reference itself, so no client-side
+resampling is required.
+
+## API example
+
+```bash
+curl http://localhost:8080/v1/audio/speech \
+  -H 'Content-Type: application/json' \
+  -d '{
+    "model": "moss-tts-cpp",
+    "input": "This request uses a saved reference voice.",
+    "voice": "/path/to/reference.wav"
+  }' \
+  --output speech.wav
+```
+
+## Native end-to-end test
+
+The labeled test loads real GGUFs, synthesizes a 48 kHz stereo WAV, and streams
+audio:
+
+```bash
+make -C backend/go/moss-tts-cpp moss-tts-cpp
+
+MOSSTTS_MODEL=/path/to/moss-tts-local-v1_5-q8_0.gguf \
+MOSSTTS_CODEC=/path/to/moss-audio-tokenizer-v2-f32.gguf \
+MOSSTTS_TOKENIZER=/path/to/moss-tokenizer-v1_5.gguf \
+MOSSTTS_LIBRARY=backend/go/moss-tts-cpp/libgomosstts-cpp-fallback.so \
+  go test ./backend/go/moss-tts-cpp -ginkgo.label-filter=e2e
+```
--- a/backend/go/moss-tts-cpp/audio.go
+++ b/backend/go/moss-tts-cpp/audio.go
@@ -0,0 +1,100 @@
+package main
+
+import (
+	"bytes"
+	"encoding/binary"
+	"fmt"
+	"os"
+
+	"github.com/go-audio/audio"
+	"github.com/go-audio/wav"
+)
+
+// mossttsSampleRate is the MOSS-TTS-Local v1.5 output rate; the engine also
+// reports it per-synthesis via out_sr, which the writers honour.
+const mossttsSampleRate = 48000
+
+// mossttsChannels is the MOSS-TTS-Local output layout: 48 kHz stereo, samples
+// interleaved [L,R,L,R,...].
+const mossttsChannels = 2
+
+// wavHeaderStereo returns a 44-byte WAV header for a streaming 16-bit stereo PCM
+// stream at sampleRate, with placeholder (0xFFFFFFFF) sizes since the total
+// length is unknown up front. Emitted as the first chunk of TTSStream so the
+// HTTP layer receives a self-describing WAV.
+func wavHeaderStereo(sampleRate int) []byte {
+	if sampleRate <= 0 {
+		sampleRate = mossttsSampleRate
+	}
+	const blockAlign = mossttsChannels * 2 // 16-bit stereo
+	var buf bytes.Buffer
+	w := func(v any) { _ = binary.Write(&buf, binary.LittleEndian, v) }
+	buf.WriteString("RIFF")
+	w(uint32(0xFFFFFFFF))
+	buf.WriteString("WAVE")
+	buf.WriteString("fmt ")
+	w(uint32(16))                      // Subchunk1Size
+	w(uint16(1))                       // PCM
+	w(uint16(mossttsChannels))         // stereo
+	w(uint32(sampleRate))              // sample rate
+	w(uint32(sampleRate * blockAlign)) // byte rate = SR * blockAlign
+	w(uint16(blockAlign))              // block align
+	w(uint16(16))                      // bits per sample
+	buf.WriteString("data")
+	w(uint32(0xFFFFFFFF))
+	return buf.Bytes()
+}
+
+// floatToPCM16LE clamps each sample to [-1,1] and encodes it as little-endian
+// signed 16-bit PCM. Input may be interleaved stereo; the layout is preserved.
+func floatToPCM16LE(samples []float32) []byte {
+	out := make([]byte, len(samples)*2)
+	for i, s := range samples {
+		if s > 1 {
+			s = 1
+		} else if s < -1 {
+			s = -1
+		}
+		v := int16(s * 32767)
+		out[i*2] = byte(v)        // #nosec G115 -- intentional little-endian split of a clamped int16
+		out[i*2+1] = byte(v >> 8) // #nosec G115 -- high byte of the same clamped int16
+	}
+	return out
+}
+
+// writeWAVStereo writes interleaved [L,R,...] float samples as a finalized
+// 16-bit stereo WAV at sampleRate (falling back to the v1.5 default of 48 kHz).
+func writeWAVStereo(dst string, samples []float32, sampleRate int) error {
+	if sampleRate <= 0 {
+		sampleRate = mossttsSampleRate
+	}
+	f, err := os.Create(dst) // #nosec G304 -- dst is the server-chosen output path from the TTS request, not user-traversable
+	if err != nil {
+		return fmt.Errorf("moss-tts: create %q: %w", dst, err)
+	}
+	enc := wav.NewEncoder(f, sampleRate, 16, mossttsChannels, 1)
+	ints := make([]int, len(samples))
+	for i, s := range samples {
+		if s > 1 {
+			s = 1
+		} else if s < -1 {
+			s = -1
+		}
+		ints[i] = int(s * 32767)
+	}
+	b := &audio.IntBuffer{
+		Format:         &audio.Format{NumChannels: mossttsChannels, SampleRate: sampleRate},
+		Data:           ints,
+		SourceBitDepth: 16,
+	}
+	if err := enc.Write(b); err != nil {
+		_ = enc.Close()
+		_ = f.Close()
+		return fmt.Errorf("moss-tts: encode WAV: %w", err)
+	}
+	if err := enc.Close(); err != nil {
+		_ = f.Close()
+		return fmt.Errorf("moss-tts: finalize WAV: %w", err)
+	}
+	return f.Close()
+}
--- a/backend/go/moss-tts-cpp/audiopath_test.go
+++ b/backend/go/moss-tts-cpp/audiopath_test.go
@@ -0,0 +1,49 @@
+package main
+
+import (
+	"path/filepath"
+
+	pb "github.com/mudler/LocalAI/pkg/grpc/proto"
+	. "github.com/onsi/ginkgo/v2"
+	. "github.com/onsi/gomega"
+)
+
+// These specs pin the voice-selection logic in resolveRequest, in particular
+// the config-level audio_path (tts.audio_path -> ModelOptions.AudioPath) being
+// used as the default voice-cloning reference. MOSS-TTS-Local takes the
+// reference as a WAV path (the engine decodes it), so resolveRequest is pure
+// path logic and needs no model / C library.
+var _ = Describe("resolveRequest voice/clone selection", func() {
+	var dir, refWav string
+
+	BeforeEach(func() {
+		dir = GinkgoT().TempDir()
+		refWav = filepath.Join(dir, "ref.wav")
+	})
+
+	It("uses the config audio_path as the clone reference when Voice is empty", func() {
+		m := &MossTtsCpp{audioPath: refWav}
+		ref, seed := m.resolveRequest(&pb.TTSRequest{Text: "hi"})
+		Expect(ref).To(Equal(refWav))
+		Expect(seed).To(Equal(0)) // zero-value loadOptions seed
+	})
+
+	It("lets a per-request audio Voice override audio_path", func() {
+		other := filepath.Join(dir, "other.wav")
+		m := &MossTtsCpp{audioPath: refWav}
+		ref, _ := m.resolveRequest(&pb.TTSRequest{Text: "hi", Voice: other})
+		Expect(ref).To(Equal(other))
+	})
+
+	It("does not clone for a bare-token Voice, falling back to audio_path", func() {
+		m := &MossTtsCpp{audioPath: refWav}
+		ref, _ := m.resolveRequest(&pb.TTSRequest{Text: "hi", Voice: "serena"})
+		Expect(ref).To(Equal(refWav))
+	})
+
+	It("reads a per-request seed override from params", func() {
+		m := &MossTtsCpp{opts: loadOptions{seed: -1}}
+		_, seed := m.resolveRequest(&pb.TTSRequest{Text: "hi", Params: map[string]string{"seed": "123"}})
+		Expect(seed).To(Equal(123))
+	})
+})
--- a/backend/go/moss-tts-cpp/cpp/mossttscpp.cpp
+++ b/backend/go/moss-tts-cpp/cpp/mossttscpp.cpp
@@ -0,0 +1,119 @@
+#include "mossttscpp.h"
+#include "ggml-backend.h"
+#include "moss_tts_capi.h"
+
+#include <cstdio>
+#include <cstdlib>
+#include <cstring>
+
+// Single loaded Local pipeline. base.SingleThread serializes calls on the Go
+// side, so a plain global is safe (mirrors qwen3-tts-cpp's g_ctx).
+static moss_local *g_local = nullptr;
+
+static void ggml_log_cb(enum ggml_log_level level, const char *log,
+                        void * /*data*/) {
+    if (!log)
+        return;
+    const char *lvl = "?????";
+    switch (level) {
+    case GGML_LOG_LEVEL_DEBUG: lvl = "DEBUG"; break;
+    case GGML_LOG_LEVEL_INFO:  lvl = "INFO";  break;
+    case GGML_LOG_LEVEL_WARN:  lvl = "WARN";  break;
+    case GGML_LOG_LEVEL_ERROR: lvl = "ERROR"; break;
+    default: break;
+    }
+    fprintf(stderr, "[%-5s] %s", lvl, log);
+    fflush(stderr);
+}
+
+int mtl_load(const char *local_path, const char *codec_path,
+             const char *tokenizer_path) {
+    ggml_log_set(ggml_log_cb, nullptr);
+    ggml_backend_load_all();
+
+    if (!local_path || local_path[0] == '\0') {
+        fprintf(stderr, "[moss-tts-cpp] ERROR: local_path is required\n");
+        return 1;
+    }
+    if (!codec_path || codec_path[0] == '\0') {
+        fprintf(stderr, "[moss-tts-cpp] ERROR: codec_path is required\n");
+        return 2;
+    }
+    if (!tokenizer_path || tokenizer_path[0] == '\0') {
+        fprintf(stderr, "[moss-tts-cpp] ERROR: tokenizer_path is required\n");
+        return 3;
+    }
+
+    fprintf(stderr,
+            "[moss-tts-cpp] Loading local=%s codec=%s tokenizer=%s\n",
+            local_path, codec_path, tokenizer_path);
+
+    g_local = moss_local_load(local_path, codec_path, tokenizer_path);
+    if (!g_local) {
+        fprintf(stderr, "[moss-tts-cpp] FATAL: moss_local_load failed\n");
+        return 4;
+    }
+    fprintf(stderr, "[moss-tts-cpp] Model loaded (%s)\n", moss_tts_version());
+    return 0;
+}
+
+float *mtl_tts(const char *text, const char *reference_wav, int seed,
+               int *out_n, int *out_sr) {
+    if (out_n)
+        *out_n = 0;
+    if (out_sr)
+        *out_sr = 0;
+    if (!g_local) {
+        fprintf(stderr, "[moss-tts-cpp] ERROR: model not loaded\n");
+        return nullptr;
+    }
+    if (!text || text[0] == '\0') {
+        fprintf(stderr, "[moss-tts-cpp] ERROR: text is required\n");
+        return nullptr;
+    }
+
+    // An empty reference path means "no cloning": pass NULL so the engine skips
+    // the clone branch rather than trying to open "".
+    const char *ref = (reference_wav && reference_wav[0] != '\0')
+                          ? reference_wav
+                          : nullptr;
+
+    int n = 0, sr = 0;
+    float *pcm = moss_local_tts(g_local, text, ref, seed, &n, &sr);
+    if (!pcm || n <= 0) {
+        fprintf(stderr, "[moss-tts-cpp] ERROR: moss_local_tts failed\n");
+        if (pcm)
+            moss_free(pcm);
+        return nullptr;
+    }
+
+    // Copy into a plain malloc buffer the Go side frees via mtl_pcm_free, then
+    // release the engine-owned buffer with moss_free (mirrors qwen3-tts-cpp,
+    // keeping ownership on the C runtime's malloc/free).
+    size_t bytes = (size_t)n * sizeof(float);
+    float *buf = (float *)malloc(bytes);
+    if (!buf) {
+        fprintf(stderr, "[moss-tts-cpp] ERROR: malloc(%zu) failed\n", bytes);
+        moss_free(pcm);
+        return nullptr;
+    }
+    memcpy(buf, pcm, bytes);
+    moss_free(pcm);
+
+    if (out_n)
+        *out_n = n;
+    if (out_sr)
+        *out_sr = sr;
+    return buf;
+}
+
+void mtl_pcm_free(float *p) { free(p); }
+
+void mtl_unload(void) {
+    if (g_local) {
+        moss_local_free(g_local);
+        g_local = nullptr;
+    }
+}
+
+const char *mtl_version(void) { return moss_tts_version(); }
--- a/backend/go/moss-tts-cpp/cpp/mossttscpp.h
+++ b/backend/go/moss-tts-cpp/cpp/mossttscpp.h
@@ -0,0 +1,36 @@
+#pragma once
+
+// Thin C shim over moss-tts.cpp's flat C-API (include/moss_tts_capi.h) for the
+// MOSS-TTS-Local (v1.5) pipeline. It holds the loaded pipeline as a single
+// global handle so the purego bindings on the Go side stay handle-free (mirrors
+// the qwen3-tts-cpp qt3_* shim). Only the Local variant is wired here; the
+// Delay / Realtime / Nano variants of the upstream API are intentionally
+// unused.
+
+extern "C" {
+
+// Load the Local (MossTTSLocal v1.5) pipeline from its three GGUFs: the local
+// transformer, the MOSS-Audio-Tokenizer codec, and the text tokenizer. Sets up
+// ggml logging + backend registration first. Returns 0 on success, non-zero on
+// failure.
+int mtl_load(const char *local_path, const char *codec_path,
+             const char *tokenizer_path);
+
+// Synthesize `text` to a malloc'd interleaved f32 PCM buffer (caller frees via
+// mtl_pcm_free). v1.5 output is 48 kHz stereo: *out_n is the total number of
+// float samples across both channels (frames * 2) and *out_sr is the sample
+// rate. `reference_wav` is an optional path to a clone-reference WAV (may be
+// NULL/empty for no cloning); the engine decodes it itself. `seed` < 0 means
+// random. Returns NULL on failure (out_n / out_sr set to 0).
+float *mtl_tts(const char *text, const char *reference_wav, int seed,
+               int *out_n, int *out_sr);
+
+// Free a buffer returned by mtl_tts.
+void mtl_pcm_free(float *p);
+
+// Release the loaded pipeline.
+void mtl_unload(void);
+
+// Upstream engine version string (moss_tts_version()).
+const char *mtl_version(void);
+}
--- a/backend/go/moss-tts-cpp/e2e_test.go
+++ b/backend/go/moss-tts-cpp/e2e_test.go
@@ -0,0 +1,72 @@
+package main
+
+import (
+	"os"
+	"strings"
+
+	"github.com/ebitengine/purego"
+	pb "github.com/mudler/LocalAI/pkg/grpc/proto"
+	. "github.com/onsi/ginkgo/v2"
+	. "github.com/onsi/gomega"
+)
+
+func ttsReq(text, voice, dst string) *pb.TTSRequest {
+	return &pb.TTSRequest{Text: text, Voice: voice, Dst: dst}
+}
+
+var _ = Describe("moss-tts-cpp e2e", Label("e2e"), func() {
+	var loaded bool
+
+	BeforeEach(func() {
+		modelPath := os.Getenv("MOSSTTS_MODEL")
+		codecPath := os.Getenv("MOSSTTS_CODEC")
+		tokenizerPath := os.Getenv("MOSSTTS_TOKENIZER")
+		if modelPath == "" || codecPath == "" || tokenizerPath == "" {
+			Skip("MOSSTTS_MODEL / MOSSTTS_CODEC / MOSSTTS_TOKENIZER not set; skipping e2e")
+		}
+		if !loaded {
+			lib := os.Getenv("MOSSTTS_LIBRARY")
+			if lib == "" {
+				lib = "./libgomosstts-cpp-fallback.so"
+			}
+			h, err := purego.Dlopen(lib, purego.RTLD_NOW|purego.RTLD_GLOBAL)
+			Expect(err).ToNot(HaveOccurred())
+			purego.RegisterLibFunc(&CppLoad, h, "mtl_load")
+			purego.RegisterLibFunc(&CppTTS, h, "mtl_tts")
+			purego.RegisterLibFunc(&CppPCMFree, h, "mtl_pcm_free")
+			purego.RegisterLibFunc(&CppUnload, h, "mtl_unload")
+			Expect(CppLoad(modelPath, codecPath, tokenizerPath)).To(Equal(0))
+			loaded = true
+		}
+	})
+
+	It("synthesizes a stereo WAV file via TTS", func() {
+		m := &MossTtsCpp{opts: loadOptions{seed: 42}}
+		dst := GinkgoT().TempDir() + "/out.wav"
+		err := m.TTS(ttsReq("Hello world.", "", dst))
+		Expect(err).ToNot(HaveOccurred())
+		fi, err := os.Stat(dst)
+		Expect(err).ToNot(HaveOccurred())
+		Expect(fi.Size()).To(BeNumerically(">", int64(44)))
+	})
+
+	It("streams audio chunks via TTSStream", func() {
+		m := &MossTtsCpp{opts: loadOptions{seed: 42}}
+		results := make(chan []byte, 4096)
+		done := make(chan error, 1)
+		go func() { done <- m.TTSStream(ttsReq("Hello there, streaming test.", "", ""), results) }()
+
+		var chunks int
+		var first []byte
+		for c := range results {
+			if chunks == 0 {
+				first = c
+			}
+			chunks++
+		}
+		Expect(<-done).ToNot(HaveOccurred())
+		Expect(chunks).To(BeNumerically(">=", 2))
+		Expect(string(first[0:4])).To(Equal("RIFF"))
+		Expect(strings.HasPrefix(string(first[8:12]), "WAVE")).To(BeTrue())
+	})
+})
--- a/backend/go/moss-tts-cpp/gomossttscpp.go
+++ b/backend/go/moss-tts-cpp/gomossttscpp.go
@@ -0,0 +1,224 @@
+package main
+
+import (
+	"fmt"
+	"os"
+	"path/filepath"
+	"strings"
+	"unsafe"
+
+	"github.com/mudler/LocalAI/pkg/grpc/base"
+	pb "github.com/mudler/LocalAI/pkg/grpc/proto"
+	"github.com/mudler/xlog"
+)
+
+var (
+	// mtl_load(local_path, codec_path, tokenizer_path) int
+	CppLoad func(localPath, codecPath, tokenizerPath string) int
+	// mtl_tts(text, reference_wav, seed, out_n, out_sr) -> float*
+	CppTTS func(text, referenceWav string, seed int, outN, outSR unsafe.Pointer) uintptr
+	// mtl_pcm_free(ptr)
+	CppPCMFree func(ptr uintptr)
+	// mtl_unload()
+	CppUnload func()
+)
+
+type MossTtsCpp struct {
+	base.SingleThread
+	opts loadOptions
+	// audioPath is the model-config reference voice (tts.audio_path), the
+	// default clone reference when a request omits an audio Voice.
+	audioPath string
+}
+
+func (m *MossTtsCpp) Load(opts *pb.ModelOptions) error {
+	model := opts.ModelFile
+	if model == "" {
+		model = opts.ModelPath
+	}
+	if !filepath.IsAbs(model) && opts.ModelPath != "" {
+		model = filepath.Join(opts.ModelPath, model)
+	}
+
+	m.opts = parseOptions(opts.Options)
+	dir := filepath.Dir(model)
+
+	// Resolve the codec GGUF (MOSS-Audio-Tokenizer): explicit option, else
+	// auto-discover an *audio*tokenizer*/codec sibling of the model.
+	codec := resolveAux(m.opts.codecPath, dir)
+	if codec == "" {
+		codec = discoverCodec(dir, model)
+	}
+	if codec == "" {
+		return fmt.Errorf("moss-tts: no codec GGUF found; set option 'codec:<file>'")
+	}
+	m.opts.codecPath = codec
+
+	// Resolve the text tokenizer GGUF: explicit option, else auto-discover the
+	// *tokenizer* sibling that is not the audio codec or the model.
+	tokenizer := resolveAux(m.opts.tokenizerPath, dir)
+	if tokenizer == "" {
+		tokenizer = discoverTokenizer(dir, model, codec)
+	}
+	if tokenizer == "" {
+		return fmt.Errorf("moss-tts: no tokenizer GGUF found; set option 'tokenizer:<file>'")
+	}
+	m.opts.tokenizerPath = tokenizer
+
+	m.audioPath = opts.AudioPath
+	if m.audioPath != "" && !filepath.IsAbs(m.audioPath) {
+		m.audioPath = filepath.Join(dir, m.audioPath)
+	}
+
+	xlog.Info("[moss-tts-cpp] Load", "model", model, "codec", codec, "tokenizer", tokenizer)
+
+	if rc := CppLoad(model, codec, tokenizer); rc != 0 {
+		return fmt.Errorf("moss-tts: failed to load model (rc=%d)", rc)
+	}
+	return nil
+}
+
+// resolveAux resolves an explicitly configured auxiliary GGUF path relative to
+// the model directory when it is not absolute. Empty stays empty.
+func resolveAux(p, dir string) string {
+	if p == "" || filepath.IsAbs(p) {
+		return p
+	}
+	return filepath.Join(dir, p)
+}
+
+// isAudioCodecName reports whether a GGUF filename denotes the MOSS-Audio-
+// Tokenizer codec (e.g. moss-audio-tokenizer-v2-f32.gguf) rather than the text
+// tokenizer (moss-tokenizer-v1_5.gguf).
+func isAudioCodecName(name string) bool {
+	n := strings.ToLower(name)
+	return strings.Contains(n, "codec") ||
+		(strings.Contains(n, "audio") && strings.Contains(n, "tokenizer"))
+}
+
+// discoverCodec returns the first codec GGUF in dir (excluding the model), or "".
+func discoverCodec(dir, model string) string {
+	entries, err := os.ReadDir(dir)
+	if err != nil {
+		return ""
+	}
+	modelBase := filepath.Base(model)
+	for _, e := range entries {
+		name := e.Name()
+		if name == modelBase || !strings.HasSuffix(strings.ToLower(name), ".gguf") {
+			continue
+		}
+		if isAudioCodecName(name) {
+			return filepath.Join(dir, name)
+		}
+	}
+	return ""
+}
+
+// discoverTokenizer returns the first text-tokenizer GGUF in dir that is neither
+// the model nor the audio codec, or "".
+func discoverTokenizer(dir, model, codec string) string {
+	entries, err := os.ReadDir(dir)
+	if err != nil {
+		return ""
+	}
+	modelBase := filepath.Base(model)
+	codecBase := filepath.Base(codec)
+	for _, e := range entries {
+		name := e.Name()
+		lower := strings.ToLower(name)
+		if name == modelBase || name == codecBase || !strings.HasSuffix(lower, ".gguf") {
+			continue
+		}
+		if strings.Contains(lower, "tokenizer") && !isAudioCodecName(name) {
+			return filepath.Join(dir, name)
+		}
+	}
+	return ""
+}
+
+// resolveRequest derives the synthesis inputs from a TTSRequest: the optional
+// clone-reference WAV path and the seed. MOSS-TTS-Local drives cloning purely
+// from a reference-audio path (the engine decodes it), so there is no
+// language / speaker / instruct plumbing here (unlike qwen3-tts-cpp).
+func (m *MossTtsCpp) resolveRequest(req *pb.TTSRequest) (refPath string, seed int) {
+	refPath = resolveVoice(req.Voice)
+	if refPath == "" && m.audioPath != "" {
+		// No per-request voice: fall back to the config clone reference.
+		refPath = m.audioPath
+	}
+	if refPath != "" && !filepath.IsAbs(refPath) {
+		refPath = filepath.Join(filepath.Dir(m.opts.codecPath), refPath)
+	}
+
+	seed = m.opts.seed
+	if req.Params != nil {
+		seed = parseInt(req.Params["seed"], seed)
+	}
+	return
+}
+
+func (m *MossTtsCpp) TTS(req *pb.TTSRequest) error {
+	if req.Dst == "" {
+		return fmt.Errorf("moss-tts: TTS requires a destination path")
+	}
+	if req.Text == "" {
+		return fmt.Errorf("moss-tts: TTS requires text")
+	}
+	refPath, seed := m.resolveRequest(req)
+
+	var n, sr int32
+	ptr := CppTTS(req.Text, refPath, seed, unsafe.Pointer(&n), unsafe.Pointer(&sr)) // #nosec G103 -- out-params for the purego-bound C-API
+	if ptr == 0 {
+		return fmt.Errorf("moss-tts: synthesis failed")
+	}
+	// Register the free as soon as we own a non-null buffer, so the n<=0 guard
+	// below cannot leak it (defensive: the C contract returns NULL on failure).
+	defer CppPCMFree(ptr)
+	if n <= 0 {
+		return fmt.Errorf("moss-tts: synthesis produced no samples")
+	}
+	//nolint:govet // C-allocated PCM, copied out before free
+	src := unsafe.Slice((*float32)(unsafe.Pointer(ptr)), int(n)) // #nosec G103 -- C-allocated PCM, copied out before free
+	out := make([]float32, int(n))
+	copy(out, src)
+	return writeWAVStereo(req.Dst, out, int(sr))
+}
+
+// TTSStream synthesizes one-shot (MOSS-TTS-Local has no native streaming C-API)
+// and then emits a self-describing stereo WAV: a header chunk followed by the
+// interleaved PCM in fixed-size slices, so the HTTP layer still receives a
+// streamed WAV (the gRPC TTSStream path never sets Message, so the backend owns
+// the header - see core/backend/tts.go:ModelTTSStream).
+func (m *MossTtsCpp) TTSStream(req *pb.TTSRequest, results chan []byte) error {
+	defer close(results)
+	if req.Text == "" {
+		return fmt.Errorf("moss-tts: TTSStream requires text")
+	}
+	refPath, seed := m.resolveRequest(req)
+
+	var n, sr int32
+	ptr := CppTTS(req.Text, refPath, seed, unsafe.Pointer(&n), unsafe.Pointer(&sr)) // #nosec G103 -- out-params for the purego-bound C-API
+	if ptr == 0 {
+		return fmt.Errorf("moss-tts: synthesis failed")
+	}
+	defer CppPCMFree(ptr)
+	if n <= 0 {
+		return fmt.Errorf("moss-tts: synthesis produced no samples")
+	}
+	//nolint:govet // C-allocated PCM, copied out before free
+	src := unsafe.Slice((*float32)(unsafe.Pointer(ptr)), int(n)) // #nosec G103 -- C-allocated PCM, copied out before free
+	out := make([]float32, int(n))
+	copy(out, src)
+
+	results <- wavHeaderStereo(int(sr))
+	const frameChunk = 4096 // interleaved stereo samples per emitted chunk
+	for off := 0; off < len(out); off += frameChunk {
+		end := off + frameChunk
+		if end > len(out) {
+			end = len(out)
+		}
+		results <- floatToPCM16LE(out[off:end])
+	}
+	return nil
+}
--- a/backend/go/moss-tts-cpp/main.go
+++ b/backend/go/moss-tts-cpp/main.go
@@ -0,0 +1,52 @@
+package main
+
+// Note: this is started internally by LocalAI and a server is allocated for each model
+import (
+	"flag"
+	"os"
+	"runtime"
+
+	"github.com/ebitengine/purego"
+	grpc "github.com/mudler/LocalAI/pkg/grpc"
+)
+
+var (
+	addr = flag.String("addr", "localhost:50051", "the address to connect to")
+)
+
+type LibFuncs struct {
+	FuncPtr any
+	Name    string
+}
+
+func main() {
+	libName := os.Getenv("MOSSTTS_LIBRARY")
+	if libName == "" {
+		if runtime.GOOS == "darwin" {
+			libName = "./libgomosstts-cpp-fallback.dylib"
+		} else {
+			libName = "./libgomosstts-cpp-fallback.so"
+		}
+	}
+
+	lib, err := purego.Dlopen(libName, purego.RTLD_NOW|purego.RTLD_GLOBAL)
+	if err != nil {
+		panic(err)
+	}
+
+	libFuncs := []LibFuncs{
+		{&CppLoad, "mtl_load"},
+		{&CppTTS, "mtl_tts"},
+		{&CppPCMFree, "mtl_pcm_free"},
+		{&CppUnload, "mtl_unload"},
+	}
+	for _, lf := range libFuncs {
+		purego.RegisterLibFunc(lf.FuncPtr, lib, lf.Name)
+	}
+
+	flag.Parse()
+
+	if err := grpc.StartServer(*addr, &MossTtsCpp{}); err != nil {
+		panic(err)
+	}
+}
--- a/backend/go/moss-tts-cpp/mossttscpp_test.go
+++ b/backend/go/moss-tts-cpp/mossttscpp_test.go
@@ -0,0 +1,124 @@
+package main
+
+import (
+	"bytes"
+	"encoding/binary"
+	"os"
+	"path/filepath"
+	"testing"
+
+	. "github.com/onsi/ginkgo/v2"
+	. "github.com/onsi/gomega"
+)
+
+func TestMossTtsCpp(t *testing.T) {
+	RegisterFailHandler(Fail)
+	RunSpecs(t, "moss-tts-cpp suite")
+}
+
+var _ = Describe("resolveVoice", func() {
+	It("treats an audio path as a clone reference (case-insensitive ext)", func() {
+		Expect(resolveVoice("/x/ref.WAV")).To(Equal("/x/ref.WAV"))
+	})
+	It("recognizes mp3/flac/ogg/m4a", func() {
+		for _, p := range []string{"a.mp3", "b.flac", "c.ogg", "d.m4a"} {
+			Expect(resolveVoice(p)).To(Equal(p))
+		}
+	})
+	It("ignores a bare token (no named speakers in MOSS-TTS-Local)", func() {
+		Expect(resolveVoice("serena")).To(BeEmpty())
+	})
+	It("returns empty for empty input", func() {
+		Expect(resolveVoice("  ")).To(BeEmpty())
+	})
+})
+
+var _ = Describe("parseOptions", func() {
+	It("extracts codec, tokenizer, seed", func() {
+		o := parseOptions([]string{
+			"codec:codec.gguf", "tokenizer:tok.gguf", "seed:7", "unknown:ignored",
+		})
+		Expect(o.codecPath).To(Equal("codec.gguf"))
+		Expect(o.tokenizerPath).To(Equal("tok.gguf"))
+		Expect(o.seed).To(Equal(7))
+	})
+	It("accepts audio_tokenizer / text_tokenizer aliases", func() {
+		o := parseOptions([]string{"audio_tokenizer:c.gguf", "text_tokenizer:t.gguf"})
+		Expect(o.codecPath).To(Equal("c.gguf"))
+		Expect(o.tokenizerPath).To(Equal("t.gguf"))
+	})
+	It("defaults seed -1", func() {
+		Expect(parseOptions(nil).seed).To(Equal(-1))
+	})
+})
+
+var _ = Describe("isAudioCodecName", func() {
+	DescribeTable("distinguishes the audio codec from the text tokenizer",
+		func(name string, want bool) {
+			Expect(isAudioCodecName(name)).To(Equal(want))
+		},
+		Entry("audio tokenizer is the codec", "moss-audio-tokenizer-v2-f32.gguf", true),
+		Entry("codec keyword", "some-codec.gguf", true),
+		Entry("text tokenizer is not the codec", "moss-tokenizer-v1_5.gguf", false),
+		Entry("model is not the codec", "moss-tts-local-v1_5-q8_0.gguf", false),
+	)
+})
+
+var _ = Describe("discoverCodec / discoverTokenizer", func() {
+	var dir, model, codec, tok string
+	BeforeEach(func() {
+		dir = GinkgoT().TempDir()
+		model = filepath.Join(dir, "moss-tts-local-v1_5-q8_0.gguf")
+		codec = filepath.Join(dir, "moss-audio-tokenizer-v2-f32.gguf")
+		tok = filepath.Join(dir, "moss-tokenizer-v1_5.gguf")
+		for _, p := range []string{model, codec, tok} {
+			Expect(os.WriteFile(p, []byte("x"), 0o644)).To(Succeed())
+		}
+	})
+	It("finds the audio codec, excluding the model", func() {
+		Expect(discoverCodec(dir, model)).To(Equal(codec))
+	})
+	It("finds the text tokenizer, excluding model and codec", func() {
+		Expect(discoverTokenizer(dir, model, codec)).To(Equal(tok))
+	})
+})
+
+var _ = Describe("wavHeaderStereo", func() {
+	It("emits a 44-byte streaming WAV header at 48 kHz stereo 16-bit", func() {
+		h := wavHeaderStereo(48000)
+		Expect(h).To(HaveLen(44))
+		Expect(string(h[0:4])).To(Equal("RIFF"))
+		Expect(string(h[8:12])).To(Equal("WAVE"))
+		Expect(string(h[12:16])).To(Equal("fmt "))
+		Expect(string(h[36:40])).To(Equal("data"))
+		var channels uint16
+		Expect(binary.Read(bytes.NewReader(h[22:24]), binary.LittleEndian, &channels)).To(Succeed())
+		Expect(channels).To(Equal(uint16(2)))
+		var sampleRate uint32
+		Expect(binary.Read(bytes.NewReader(h[24:28]), binary.LittleEndian, &sampleRate)).To(Succeed())
+		Expect(sampleRate).To(Equal(uint32(48000)))
+	})
+	It("falls back to 48 kHz for a non-positive sample rate", func() {
+		var sampleRate uint32
+		h := wavHeaderStereo(0)
+		Expect(binary.Read(bytes.NewReader(h[24:28]), binary.LittleEndian, &sampleRate)).To(Succeed())
+		Expect(sampleRate).To(Equal(uint32(48000)))
+	})
+})
+
+var _ = Describe("floatToPCM16LE", func() {
+	It("clamps and converts float PCM to little-endian int16 bytes", func() {
+		b := floatToPCM16LE([]float32{0, 1.0, -1.0, 2.0, -2.0})
+		Expect(b).To(HaveLen(10))
+		read := func(off int) int16 {
+			var v int16
+			_ = binary.Read(bytes.NewReader(b[off:off+2]), binary.LittleEndian, &v)
+			return v
+		}
+		Expect(read(0)).To(Equal(int16(0)))
+		Expect(read(2)).To(Equal(int16(32767)))
+		Expect(read(4)).To(Equal(int16(-32767)))
+		Expect(read(6)).To(Equal(int16(32767)))  // clamped from 2.0
+		Expect(read(8)).To(Equal(int16(-32767))) // clamped from -2.0
+	})
+})
--- a/backend/go/moss-tts-cpp/options.go
+++ b/backend/go/moss-tts-cpp/options.go
@@ -0,0 +1,78 @@
+package main
+
+import (
+	"strconv"
+	"strings"
+)
+
+// loadOptions holds the parsed model-level options. MOSS-TTS-Local needs three
+// GGUFs (local transformer, audio codec, text tokenizer); the model path is the
+// local transformer and codec/tokenizer are resolved as siblings or via these
+// options.
+type loadOptions struct {
+	codecPath     string
+	tokenizerPath string
+	seed          int
+}
+
+func splitOption(o string) (key, value string, ok bool) {
+	i := strings.Index(o, ":")
+	if i < 0 {
+		return "", "", false
+	}
+	return strings.TrimSpace(o[:i]), strings.TrimSpace(o[i+1:]), true
+}
+
+// parseOptions reads the backend "key:value" option slice. Unknown keys are
+// ignored. Default seed is -1 (engine random).
+func parseOptions(opts []string) loadOptions {
+	o := loadOptions{seed: -1}
+	for _, oo := range opts {
+		key, value, ok := splitOption(oo)
+		if !ok {
+			continue
+		}
+		switch key {
+		case "codec", "audio_tokenizer":
+			o.codecPath = value
+		case "tokenizer", "text_tokenizer":
+			o.tokenizerPath = value
+		case "seed":
+			if n, err := strconv.Atoi(value); err == nil {
+				o.seed = n
+			}
+		}
+	}
+	return o
+}
+
+var refAudioExts = []string{".wav", ".flac", ".mp3", ".ogg", ".m4a"}
+
+// resolveVoice interprets the request Voice field. MOSS-TTS-Local has no named
+// speakers, only reference-audio cloning, so a value ending in a known audio
+// extension is treated as a clone-reference path and anything else is ignored
+// (the caller then falls back to the config audio_path).
+func resolveVoice(voice string) (refPath string) {
+	v := strings.TrimSpace(voice)
+	if v == "" {
+		return ""
+	}
+	lower := strings.ToLower(v)
+	for _, ext := range refAudioExts {
+		if strings.HasSuffix(lower, ext) {
+			return v
+		}
+	}
+	return ""
+}
+
+func parseInt(v string, def int) int {
+	if v == "" {
+		return def
+	}
+	n, err := strconv.Atoi(v)
+	if err != nil {
+		return def
+	}
+	return n
+}
--- a/backend/go/moss-tts-cpp/package.sh
+++ b/backend/go/moss-tts-cpp/package.sh
@@ -0,0 +1,65 @@
+#!/bin/bash
+
+# Script to copy the appropriate libraries based on architecture
+# This script is used in the final stage of the Dockerfile
+
+set -e
+
+CURDIR=$(dirname "$(realpath $0)")
+REPO_ROOT="${CURDIR}/../../.."
+
+# Create lib directory
+mkdir -p $CURDIR/package/lib
+
+cp -avf $CURDIR/moss-tts-cpp $CURDIR/package/
+cp -fv $CURDIR/libgomosstts-cpp-*.so $CURDIR/package/ 2>/dev/null || true
+cp -fv $CURDIR/libgomosstts-cpp-*.dylib $CURDIR/package/ 2>/dev/null || true
+cp -fv $CURDIR/run.sh $CURDIR/package/
+
+# Detect architecture and copy appropriate libraries
+if [ -f "/lib64/ld-linux-x86-64.so.2" ]; then
+    # x86_64 architecture
+    echo "Detected x86_64 architecture, copying x86_64 libraries..."
+    cp -arfLv /lib64/ld-linux-x86-64.so.2 $CURDIR/package/lib/ld.so
+    cp -arfLv /lib/x86_64-linux-gnu/libc.so.6 $CURDIR/package/lib/libc.so.6
+    cp -arfLv /lib/x86_64-linux-gnu/libgcc_s.so.1 $CURDIR/package/lib/libgcc_s.so.1
+    cp -arfLv /lib/x86_64-linux-gnu/libstdc++.so.6 $CURDIR/package/lib/libstdc++.so.6
+    cp -arfLv /lib/x86_64-linux-gnu/libm.so.6 $CURDIR/package/lib/libm.so.6
+    cp -arfLv /lib/x86_64-linux-gnu/libgomp.so.1 $CURDIR/package/lib/libgomp.so.1
+    cp -arfLv /lib/x86_64-linux-gnu/libgcc_s.so.1 $CURDIR/package/lib/libgcc_s.so.1
+    cp -arfLv /lib/x86_64-linux-gnu/libstdc++.so.6 $CURDIR/package/lib/libstdc++.so.6
+    cp -arfLv /lib/x86_64-linux-gnu/libdl.so.2 $CURDIR/package/lib/libdl.so.2
+    cp -arfLv /lib/x86_64-linux-gnu/librt.so.1 $CURDIR/package/lib/librt.so.1
+    cp -arfLv /lib/x86_64-linux-gnu/libpthread.so.0 $CURDIR/package/lib/libpthread.so.0
+elif [ -f "/lib/ld-linux-aarch64.so.1" ]; then
+    # ARM64 architecture
+    echo "Detected ARM64 architecture, copying ARM64 libraries..."
+    cp -arfLv /lib/ld-linux-aarch64.so.1 $CURDIR/package/lib/ld.so
+    cp -arfLv /lib/aarch64-linux-gnu/libc.so.6 $CURDIR/package/lib/libc.so.6
+    cp -arfLv /lib/aarch64-linux-gnu/libgcc_s.so.1 $CURDIR/package/lib/libgcc_s.so.1
+    cp -arfLv /lib/aarch64-linux-gnu/libstdc++.so.6 $CURDIR/package/lib/libstdc++.so.6
+    cp -arfLv /lib/aarch64-linux-gnu/libm.so.6 $CURDIR/package/lib/libm.so.6
+    cp -arfLv /lib/aarch64-linux-gnu/libgomp.so.1 $CURDIR/package/lib/libgomp.so.1
+    cp -arfLv /lib/aarch64-linux-gnu/libgcc_s.so.1 $CURDIR/package/lib/libgcc_s.so.1
+    cp -arfLv /lib/aarch64-linux-gnu/libstdc++.so.6 $CURDIR/package/lib/libstdc++.so.6
+    cp -arfLv /lib/aarch64-linux-gnu/libdl.so.2 $CURDIR/package/lib/libdl.so.2
+    cp -arfLv /lib/aarch64-linux-gnu/librt.so.1 $CURDIR/package/lib/librt.so.1
+    cp -arfLv /lib/aarch64-linux-gnu/libpthread.so.0 $CURDIR/package/lib/libpthread.so.0
+elif [ $(uname -s) = "Darwin" ]; then
+    echo "Detected Darwin"
+else
+    echo "Error: Could not detect architecture"
+    exit 1
+fi
+
+# Package GPU libraries based on BUILD_TYPE
+GPU_LIB_SCRIPT="${REPO_ROOT}/scripts/build/package-gpu-libs.sh"
+if [ -f "$GPU_LIB_SCRIPT" ]; then
+    echo "Packaging GPU libraries for BUILD_TYPE=${BUILD_TYPE:-cpu}..."
+    source "$GPU_LIB_SCRIPT" "$CURDIR/package/lib"
+    package_gpu_libs
+fi
+
+echo "Packaging completed successfully"
+ls -liah $CURDIR/package/
+ls -liah $CURDIR/package/lib/
--- a/backend/go/moss-tts-cpp/run.sh
+++ b/backend/go/moss-tts-cpp/run.sh
@@ -0,0 +1,57 @@
+#!/bin/bash
+set -ex
+
+# Get the absolute current dir where the script is located
+CURDIR=$(dirname "$(realpath "$0")")
+
+cd /
+
+echo "CPU info:"
+if [ "$(uname)" != "Darwin" ]; then
+	grep -e "model\sname" /proc/cpuinfo | head -1
+	grep -e "flags" /proc/cpuinfo | head -1
+fi
+
+if [ "$(uname)" = "Darwin" ]; then
+	# macOS: single dylib variant (Metal or Accelerate)
+	LIBRARY="$CURDIR/libgomosstts-cpp-fallback.dylib"
+	export DYLD_LIBRARY_PATH="$CURDIR"/lib:$DYLD_LIBRARY_PATH
+else
+	LIBRARY="$CURDIR/libgomosstts-cpp-fallback.so"
+
+	if grep -q -e "\savx\s" /proc/cpuinfo ; then
+		echo "CPU:    AVX    found OK"
+		if [ -e "$CURDIR"/libgomosstts-cpp-avx.so ]; then
+			LIBRARY="$CURDIR/libgomosstts-cpp-avx.so"
+		fi
+	fi
+
+	if grep -q -e "\savx2\s" /proc/cpuinfo ; then
+		echo "CPU:    AVX2   found OK"
+		if [ -e "$CURDIR"/libgomosstts-cpp-avx2.so ]; then
+			LIBRARY="$CURDIR/libgomosstts-cpp-avx2.so"
+		fi
+	fi
+
+	# Check avx 512
+	if grep -q -e "\savx512f\s" /proc/cpuinfo ; then
+		echo "CPU:    AVX512F found OK"
+		if [ -e "$CURDIR"/libgomosstts-cpp-avx512.so ]; then
+			LIBRARY="$CURDIR/libgomosstts-cpp-avx512.so"
+		fi
+	fi
+
+	export LD_LIBRARY_PATH="$CURDIR"/lib:$LD_LIBRARY_PATH
+fi
+
+export MOSSTTS_LIBRARY=$LIBRARY
+
+# If there is a lib/ld.so, use it
+if [ -f "$CURDIR"/lib/ld.so ]; then
+	echo "Using lib/ld.so"
+	echo "Using library: $LIBRARY"
+	exec "$CURDIR"/lib/ld.so "$CURDIR"/moss-tts-cpp "$@"
+fi
+
+echo "Using library: $LIBRARY"
+exec "$CURDIR"/moss-tts-cpp "$@"
--- a/backend/go/moss-tts-cpp/test.sh
+++ b/backend/go/moss-tts-cpp/test.sh
@@ -0,0 +1,33 @@
+#!/bin/bash
+set -e
+
+CURDIR=$(dirname "$(realpath $0)")
+cd "$CURDIR"
+
+echo "Running moss-tts-cpp backend tests..."
+
+# Auto-download the Local v1.5 model + codec + tokenizer only when
+# MOSSTTS_MODEL is not set.
+if [ -z "$MOSSTTS_MODEL" ]; then
+    MODEL_DIR="./moss-tts-models"
+    mkdir -p "$MODEL_DIR"
+    REPO_ID="mudler/MOSS-TTS-Local-Transformer-v1.5-GGUF"
+    BASE_URL="https://huggingface.co/${REPO_ID}/resolve/main"
+    FILES=( "moss-tts-local-v1_5-q8_0.gguf" "moss-audio-tokenizer-v2-f32.gguf" "moss-tokenizer-v1_5.gguf" )
+    for file in "${FILES[@]}"; do
+        dest="${MODEL_DIR}/${file}"
+        if [ -f "${dest}" ]; then
+            echo "  [skip] ${file}"
+        else
+            echo "  [download] ${file}..."
+            curl -L -o "${dest}" "${BASE_URL}/${file}" --progress-bar
+        fi
+    done
+    export MOSSTTS_MODEL="${MODEL_DIR}/moss-tts-local-v1_5-q8_0.gguf"
+    export MOSSTTS_CODEC="${MODEL_DIR}/moss-audio-tokenizer-v2-f32.gguf"
+    export MOSSTTS_TOKENIZER="${MODEL_DIR}/moss-tokenizer-v1_5.gguf"
+fi
+
+go test -v -timeout 1200s .
+
+echo "All moss-tts-cpp tests passed."
--- a/backend/index.yaml
+++ b/backend/index.yaml
@@ -1129,6 +1129,48 @@
    nvidia-l4t: "nvidia-l4t-arm64-qwen3-tts-cpp"
    nvidia-l4t-cuda-12: "nvidia-l4t-arm64-qwen3-tts-cpp"
    nvidia-l4t-cuda-13: "cuda13-nvidia-l4t-arm64-qwen3-tts-cpp"
+- &mossttscpp
+  name: "moss-tts-cpp"
+  description: |
+    MOSS-TTS C++ backend using GGML (moss-tts.cpp). Native C++ text-to-speech
+    for the OpenMOSS MOSS-TTS-Local Transformer v1.5 model (a GPT-J local
+    transformer + MOSS-Audio-Tokenizer-v2 neural codec), running from a
+    self-contained GGUF with no Python at inference time. 48kHz stereo output
+    with reference-audio voice cloning.
+  urls:
+    - https://github.com/mudler/moss-tts.cpp
+    - https://huggingface.co/mudler/MOSS-TTS-Local-Transformer-v1.5-GGUF
+  tags:
+    - text-to-speech
+    - tts
+    - voice-cloning
+  alias: "moss-tts-cpp"
+  capabilities:
+    default: "cpu-moss-tts-cpp"
+    nvidia: "cuda12-moss-tts-cpp"
+    nvidia-cuda-13: "cuda13-moss-tts-cpp"
+    nvidia-cuda-12: "cuda12-moss-tts-cpp"
+    intel: "intel-sycl-f16-moss-tts-cpp"
+    metal: "metal-moss-tts-cpp"
+    amd: "rocm-moss-tts-cpp"
+    vulkan: "vulkan-moss-tts-cpp"
+    nvidia-l4t: "nvidia-l4t-arm64-moss-tts-cpp"
+    nvidia-l4t-cuda-12: "nvidia-l4t-arm64-moss-tts-cpp"
+    nvidia-l4t-cuda-13: "cuda13-nvidia-l4t-arm64-moss-tts-cpp"
+- !!merge <<: *mossttscpp
+  name: "moss-tts-cpp-development"
+  capabilities:
+    default: "cpu-moss-tts-cpp-development"
+    nvidia: "cuda12-moss-tts-cpp-development"
+    nvidia-cuda-13: "cuda13-moss-tts-cpp-development"
+    nvidia-cuda-12: "cuda12-moss-tts-cpp-development"
+    intel: "intel-sycl-f16-moss-tts-cpp-development"
+    metal: "metal-moss-tts-cpp-development"
+    amd: "rocm-moss-tts-cpp-development"
+    vulkan: "vulkan-moss-tts-cpp-development"
+    nvidia-l4t: "nvidia-l4t-arm64-moss-tts-cpp-development"
+    nvidia-l4t-cuda-12: "nvidia-l4t-arm64-moss-tts-cpp-development"
+    nvidia-l4t-cuda-13: "cuda13-nvidia-l4t-arm64-moss-tts-cpp-development"
 - &omnivoicecpp
  name: "omnivoice-cpp"
  description: |
@@ -4611,6 +4653,107 @@
  uri: "quay.io/go-skynet/local-ai-backends:master-gpu-nvidia-cuda-13-qwen3-tts-cpp"
  mirrors:
    - localai/localai-backends:master-gpu-nvidia-cuda-13-qwen3-tts-cpp
+## moss-tts-cpp
+- !!merge <<: *mossttscpp
+  name: "nvidia-l4t-arm64-moss-tts-cpp"
+  uri: "quay.io/go-skynet/local-ai-backends:latest-nvidia-l4t-arm64-moss-tts-cpp"
+  mirrors:
+    - localai/localai-backends:latest-nvidia-l4t-arm64-moss-tts-cpp
+- !!merge <<: *mossttscpp
+  name: "nvidia-l4t-arm64-moss-tts-cpp-development"
+  uri: "quay.io/go-skynet/local-ai-backends:master-nvidia-l4t-arm64-moss-tts-cpp"
+  mirrors:
+    - localai/localai-backends:master-nvidia-l4t-arm64-moss-tts-cpp
+- !!merge <<: *mossttscpp
+  name: "cuda13-nvidia-l4t-arm64-moss-tts-cpp"
+  uri: "quay.io/go-skynet/local-ai-backends:latest-nvidia-l4t-cuda-13-arm64-moss-tts-cpp"
+  mirrors:
+    - localai/localai-backends:latest-nvidia-l4t-cuda-13-arm64-moss-tts-cpp
+- !!merge <<: *mossttscpp
+  name: "cuda13-nvidia-l4t-arm64-moss-tts-cpp-development"
+  uri: "quay.io/go-skynet/local-ai-backends:master-nvidia-l4t-cuda-13-arm64-moss-tts-cpp"
+  mirrors:
+    - localai/localai-backends:master-nvidia-l4t-cuda-13-arm64-moss-tts-cpp
+- !!merge <<: *mossttscpp
+  name: "cpu-moss-tts-cpp"
+  uri: "quay.io/go-skynet/local-ai-backends:latest-cpu-moss-tts-cpp"
+  mirrors:
+    - localai/localai-backends:latest-cpu-moss-tts-cpp
+- !!merge <<: *mossttscpp
+  name: "metal-moss-tts-cpp"
+  uri: "quay.io/go-skynet/local-ai-backends:latest-metal-darwin-arm64-moss-tts-cpp"
+  mirrors:
+    - localai/localai-backends:latest-metal-darwin-arm64-moss-tts-cpp
+- !!merge <<: *mossttscpp
+  name: "metal-moss-tts-cpp-development"
+  uri: "quay.io/go-skynet/local-ai-backends:master-metal-darwin-arm64-moss-tts-cpp"
+  mirrors:
+    - localai/localai-backends:master-metal-darwin-arm64-moss-tts-cpp
+- !!merge <<: *mossttscpp
+  name: "cpu-moss-tts-cpp-development"
+  uri: "quay.io/go-skynet/local-ai-backends:master-cpu-moss-tts-cpp"
+  mirrors:
+    - localai/localai-backends:master-cpu-moss-tts-cpp
+- !!merge <<: *mossttscpp
+  name: "cuda12-moss-tts-cpp"
+  uri: "quay.io/go-skynet/local-ai-backends:latest-gpu-nvidia-cuda-12-moss-tts-cpp"
+  mirrors:
+    - localai/localai-backends:latest-gpu-nvidia-cuda-12-moss-tts-cpp
+- !!merge <<: *mossttscpp
+  name: "rocm-moss-tts-cpp"
+  uri: "quay.io/go-skynet/local-ai-backends:latest-gpu-rocm-hipblas-moss-tts-cpp"
+  mirrors:
+    - localai/localai-backends:latest-gpu-rocm-hipblas-moss-tts-cpp
+- !!merge <<: *mossttscpp
+  name: "intel-sycl-f32-moss-tts-cpp"
+  uri: "quay.io/go-skynet/local-ai-backends:latest-gpu-intel-sycl-f32-moss-tts-cpp"
+  mirrors:
+    - localai/localai-backends:latest-gpu-intel-sycl-f32-moss-tts-cpp
+- !!merge <<: *mossttscpp
+  name: "intel-sycl-f16-moss-tts-cpp"
+  uri: "quay.io/go-skynet/local-ai-backends:latest-gpu-intel-sycl-f16-moss-tts-cpp"
+  mirrors:
+    - localai/localai-backends:latest-gpu-intel-sycl-f16-moss-tts-cpp
+- !!merge <<: *mossttscpp
+  name: "vulkan-moss-tts-cpp"
+  uri: "quay.io/go-skynet/local-ai-backends:latest-gpu-vulkan-moss-tts-cpp"
+  mirrors:
+    - localai/localai-backends:latest-gpu-vulkan-moss-tts-cpp
+- !!merge <<: *mossttscpp
+  name: "vulkan-moss-tts-cpp-development"
+  uri: "quay.io/go-skynet/local-ai-backends:master-gpu-vulkan-moss-tts-cpp"
+  mirrors:
+    - localai/localai-backends:master-gpu-vulkan-moss-tts-cpp
+- !!merge <<: *mossttscpp
+  name: "cuda12-moss-tts-cpp-development"
+  uri: "quay.io/go-skynet/local-ai-backends:master-gpu-nvidia-cuda-12-moss-tts-cpp"
+  mirrors:
+    - localai/localai-backends:master-gpu-nvidia-cuda-12-moss-tts-cpp
+- !!merge <<: *mossttscpp
+  name: "rocm-moss-tts-cpp-development"
+  uri: "quay.io/go-skynet/local-ai-backends:master-gpu-rocm-hipblas-moss-tts-cpp"
+  mirrors:
+    - localai/localai-backends:master-gpu-rocm-hipblas-moss-tts-cpp
+- !!merge <<: *mossttscpp
+  name: "intel-sycl-f32-moss-tts-cpp-development"
+  uri: "quay.io/go-skynet/local-ai-backends:master-gpu-intel-sycl-f32-moss-tts-cpp"
+  mirrors:
+    - localai/localai-backends:master-gpu-intel-sycl-f32-moss-tts-cpp
+- !!merge <<: *mossttscpp
+  name: "intel-sycl-f16-moss-tts-cpp-development"
+  uri: "quay.io/go-skynet/local-ai-backends:master-gpu-intel-sycl-f16-moss-tts-cpp"
+  mirrors:
+    - localai/localai-backends:master-gpu-intel-sycl-f16-moss-tts-cpp
+- !!merge <<: *mossttscpp
+  name: "cuda13-moss-tts-cpp"
+  uri: "quay.io/go-skynet/local-ai-backends:latest-gpu-nvidia-cuda-13-moss-tts-cpp"
+  mirrors:
+    - localai/localai-backends:latest-gpu-nvidia-cuda-13-moss-tts-cpp
+- !!merge <<: *mossttscpp
+  name: "cuda13-moss-tts-cpp-development"
+  uri: "quay.io/go-skynet/local-ai-backends:master-gpu-nvidia-cuda-13-moss-tts-cpp"
+  mirrors:
+    - localai/localai-backends:master-gpu-nvidia-cuda-13-moss-tts-cpp
 ## omnivoice-cpp
 - !!merge <<: *omnivoicecpp
  name: "omnivoice-cpp-development"
--- a/docs/content/features/text-to-audio.md
+++ b/docs/content/features/text-to-audio.md
@@ -127,7 +127,7 @@ When a saved profile is selected, LocalAI supplies both its private WAV and exac

 | Backend | Automatically compatible variants |
 | --- | --- |
-| `chatterbox`, `faster-qwen3-tts`, `fish-speech`, `neutts`, `omnivoice-cpp`, `pocket-tts`, `voxcpm` | Reference-audio cloning models served by these dedicated backends. |
+| `chatterbox`, `faster-qwen3-tts`, `fish-speech`, `moss-tts-cpp`, `neutts`, `omnivoice-cpp`, `pocket-tts`, `voxcpm` | Reference-audio cloning models served by these dedicated backends. |
 | `qwen-tts`, `qwen3-tts-cpp`, `vllm-omni` | Base or VoiceClone variants. CustomVoice and VoiceDesign variants are not raw reference-audio models. |
 | `vibevoice-cpp` | 1.5B reference-WAV variants. The realtime 0.5B preset-prompt model is excluded. |
 | `coqui` | XTTS and YourTTS variants. |
--- a/docs/content/reference/compatibility-table.md
+++ b/docs/content/reference/compatibility-table.md
@@ -59,6 +59,7 @@ All backends listed here can be installed on demand from the [Backend Gallery]({
 | [Chatterbox](https://github.com/resemble-ai/chatterbox) | Production-grade TTS with emotion control | CPU, CUDA 12/13, Metal, Jetson L4T |
 | [VibeVoice](https://github.com/microsoft/VibeVoice) | Real-time TTS with voice cloning | CPU, CUDA 12/13, ROCm, Intel SYCL, Metal, Jetson L4T |
 | [vibevoice.cpp](https://github.com/mudler/vibevoice.cpp) | Native C++/GGML port of VibeVoice for TTS (voice cloning) and long-form ASR with diarization | CPU, CUDA 12/13, ROCm, Intel SYCL, Vulkan, Metal, Jetson L4T |
+| [moss-tts.cpp](https://github.com/mudler/moss-tts.cpp) | Native C++/GGML port of OpenMOSS MOSS-TTS-Local v1.5: 48 kHz stereo TTS with reference-audio voice cloning | CPU, CUDA 12/13, ROCm, Intel SYCL, Vulkan, Metal, Jetson L4T |
 | [Qwen3-TTS](https://github.com/QwenLM/Qwen3-TTS) | TTS with custom voice, voice design, and voice cloning | CPU, CUDA 12/13, ROCm, Intel SYCL, Metal, Jetson L4T |
 | [qwentts.cpp](https://github.com/ServeurpersoCom/qwentts.cpp) | Native C++/GGML Qwen3-TTS with streaming, named speakers, and voice design | CPU, CUDA 12/13, ROCm, Intel SYCL, Vulkan, Metal, Jetson L4T |
 | [OmniVoice](https://github.com/ServeurpersoCom/omnivoice.cpp) | Native C++/GGML TTS with voice cloning, voice design, and streaming | CPU, CUDA 12/13, ROCm, Intel SYCL, Vulkan, Metal, Jetson L4T |
--- a/scripts/changed-backends.js
+++ b/scripts/changed-backends.js
@@ -41,6 +41,14 @@ function inferBackendPath(item) {
  if (item.backend === "moss-transcribe-cpp") {
    return `backend/go/moss-transcribe-cpp/`;
  }
+  // moss-tts-cpp is a Go backend (Dockerfile.golang) wrapping the moss-tts.cpp
+  // ggml port via purego, living in backend/go/moss-tts-cpp/. Same
+  // explicit-branch rationale as parakeet-cpp / ced / moss-transcribe-cpp: the
+  // generic golang fallthrough would also resolve it, but this documents the
+  // mapping and guards a future dockerfile-suffix change.
+  if (item.backend === "moss-tts-cpp") {
+    return `backend/go/moss-tts-cpp/`;
+  }
  if (item.dockerfile.endsWith("golang")) {
    return `backend/go/${item.backend}/`;
  }