v2.0.3

soxoj/socid-extractorv2.0.3Jun 7, 2026by lipku

AI Summary

This release introduces a new vLLM Omni Speech Adapter (OmniTTS) and a comprehensive Web UI for managing TTS voices and testing synthesis.

Key Highlights

Added OmniTTS backend supporting various vLLM models.
Introduced a Web UI for voice management, cloning, and testing.
Added support for multi-format audio output (WAV, MP3, FLAC, etc.).

New Features

vLLM Omni Speech Adapter integration.
Web-based TTS Voice Manager.
Multi-format audio export support.

Full Release Notes

### ✨ New Features

**OmniTTS — vLLM Omni Speech Adapter**
- Added `tts/omnitts.py`, a new TTS backend that calls the vLLM Omni OpenAI-compatible speech API (`POST /v1/audio/speech`). Supports all vLLM Omni models (Qwen3-TTS, Fish Speech S2, CosyVoice3, Voxtral, VoxCPM2, MOSS-TTS-Nano) with configurable source sample rate, automatic resampling to 16 kHz, and per-message parameter overrides (voice, language, speed, instructions, task type). [usage](https://doc.livetalking.ai/docs/usage/#331-omni-tts)

**TTS Voice Manager Web UI**
- Added `web/tts/index.html` and `web/tts/index-en.html` — a full-featured voice management dashboard:
- **Voice list** — browse preset and uploaded voices from the vLLM Omni server, with one-click selection and voice deletion.
- **Voice clone upload** — upload audio samples to clone new voices with auto-generated consent ID, required reference text, and a browser-native speech recognition button for auto-transcription.
- **Speech synthesis test** — select a voice, enter text, and synthesize speech with adjustable speed, language (11 languages), output format (WAV/MP3/FLAC/AAC/Opus/PCM), and task type. Results play in-page and are downloadable.
- Added quick-link cards to the main index pages pointing to the TTS manager.

**Environment Variable Template**
- Added `.env.example` with placeholder keys for Tencent, DashScope, and Doubao services. `python-dotenv` auto-loads `.env` on startup.