632c4b6db2
refactor(backends): extract shared package-system-libs.sh from package.sh
The arch-detect-and-copy-system-libs block (Darwin rpath / x86_64 / aarch64
loader + libc/libstdc++/libgcc_s/libm/libgomp/libdl/librt/libpthread) was
inlined verbatim in 31 backend package.sh scripts. Extract it into a single
sourced scripts/build/package-system-libs.sh, the CPU-side counterpart to
scripts/build/package-gpu-libs.sh and its sourcing contract.
Consolidating the copies fixes three drift classes that had crept in:
- libgcc_s.so.1 and libstdc++.so.6 were listed twice in 9 backends
(acestep-cpp, crispasr, moss-tts-cpp, omnivoice-cpp, piper,
qwen3-tts-cpp, silero-vad, stablediffusion-ggml, whisper); the shared
script copies each once.
- libgomp.so.1 was omitted from opus. OpenMP consumers dlopen it rather
than link it, so the missing copy only failed at runtime; the shared
script always includes it.
- the Darwin @loader_path/lib rpath was applied only in piper and
silero-vad; both now pass their packaged binary to the shared script,
preserving that behavior. Every other backend passes an empty binary
path so no rpath is added, preserving its current behavior.
Each backend's pre/post packaging steps (binary copy, run.sh, ldd closure
walks, ggml variant bundling, espeak/OpenBLAS extras, the ds4 validate step)
are preserved verbatim; only the inline if/elif/else arch block is replaced
by a single source line.
Signed-off-by: supermario_leo <leo.stack@outlook.com>
Co-authored-by: localai-org-maint-bot <bot-opensource@localaisrl.com>
MOSS-TTS C++ backend
This backend runs the OpenMOSS MOSS-TTS-Local (v1.5) GGUF model through 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
*.ggufwhose name containsaudio+tokenizer(orcodec), e.g.moss-audio-tokenizer-v2-f32.gguf - tokenizer: the other
*.ggufwhose name containstokenizer, e.g.moss-tokenizer-v1_5.gguf
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:
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.
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
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:
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