GPT-SoVITS TTS

A production-ready text-to-speech skill that connects to a local GPT-SoVITS v2 Pro+ API server. Converts Chinese text to natural-sounding speech with a cloned reference voice. Designed for voice response automation, content narration, and AI voice applications.

Features

• Clean TTS pipeline: Text → GPT-SoVITS API → WAV → MP3 (128kbps, 44100Hz, mono)
• Voice cloning: Uses a pre-recorded reference audio for consistent voice output
• Configurable: API URL, timeout, TTS parameters (speed, top_k, top_p, temperature, seed)
• No GPU required: Pure CPU inference, works on any machine (approx. 5-10s per sentence)

Requirements

• GPT-SoVITS v2 Pro+ API running at http://127.0.0.1:9880 (or set GPT_SOVITS_API_URL)
• ffmpeg installed and in PATH (for WAV→MP3 conversion)
• Node.js packages: axios

Model files needed (on the API server side)

Quick Start

1. Start GPT-SoVITS API

cd /path/to/GPT-SoVITS-CPUFast
conda activate GPTSoVits
python api_v2.py -a 127.0.0.1 -p 9880

2. Set reference audio

Place a clean .wav file (10-30 seconds of the target voice) at:

voice-clone/ref_audio.wav

3. Use the skill

const tts = require('./skills/voice-clone');
const mp3 = await tts.speak("你好，欢迎使用GPT-SoVITS语音合成。", "output.mp3");
// Returns: "output.mp3"

API

speak(text, outputPath, opts?)

Returns: Promise — path to the generated MP3 file.

Environment Variables

Integration

This skill is designed to be called from automation workflows:

• Voice reply for messaging bots (WeChat, Telegram, etc.)
• Content narration for video/audio production
• Voice response for IVR systems

License

MIT