Feat: Add support for VieNeu-TTS v3 Turbo (48 kHz) Model Family#80

Merged

0xShug0 merged 16 commits into

0xShug0:mainfrom

phuocnguyen90:vietneu-tts-v3-turbo

Jul 21, 2026

Commits on Jul 20, 2026

feat: implement native C++ VieNeu-TTS v3 Turbo runner and packaging support
phuocnguyen90
committed
feat: align VieNeu-TTS prompt embedding layout and speaker anchor with Python reference
phuocnguyen90
committed
feat: support loading pre-computed speaker embedding from sidecar file for offline voice cloning
phuocnguyen90
committed
fix: align C++ speaker embedding and acoustic decoder conditioning input to match Python exactly
phuocnguyen90
committed
Fix VieNeu-TTS v3 Turbo acoustic embedding mapping and text logits early stopping
phuocnguyen90
committed
Revert file mode permission change on build_linux.sh to match upstream main
phuocnguyen90
committed
Merge upstream main and resolve conflicts in qwen_decoder.cpp by adopting structured position encoding
phuocnguyen90
committed
Update VieNeu-TTS talker to use QwenDecoderPositionEncoding::None for acoustic decoder stack
phuocnguyen90
committed
Move vietneu_tts under community_models to match project contributions policy
phuocnguyen90
committed
Address review: revert CLI changes, support speaker_embedding_file option, correct error prefixes, clean up obsolete task placeholders, and register safetensors download support
phuocnguyen90
committed
Update documentation with request-option formatting
phuocnguyen90
committed
Merge remote-tracking branch 'origin/main' into vietneu-tts-v3-turbo
phuocnguyen90
committed
Use community-scoped constant tensor cache from upstream
phuocnguyen90
committed

Commits on Jul 21, 2026