rachelvg/openvoice-v2

🎙 OpenVoice V2 — Voice Cloning & Design

OpenVoice V2 by MyShell — MIT License, free for commercial use.

Features

• 🎤 Voice Clone: Upload reference audio (10s–3min) + text → cloned speech in multiple languages
• 🎨 Voice Design: Choose gender, age, pitch → novel voice without reference audio
• 🌍 Multilingual: EN, ZH, JA, KO, FR, DE, ES, PT, RU, AR, HI, IT, NL, TR, PL, VI
• ⚡ CPU Optimized: Runs on HF Spaces free tier (cpu-basic), no GPU required

How it works

OpenVoice V2 uses a two-stage approach:

1. Base TTS (VITS-based) generates speech from text using a base speaker
2. Tone Color Converter transfers the timbre from reference audio to the generated speech

Usage

1. Select Voice Clone or Voice Design tab
2. For Clone: upload reference audio, enter text, choose language/speed
3. For Design: enter text, choose characteristics, generate
4. Wait for processing (CPU inference takes 30–120s depending on length)

Models

• Base speakers: checkpoints_v2/base_speakers/{EN,ZH,JP,KR,FR,DE,ES,PT}/
• Converter: checkpoints_v2/converter/
• Auto-downloaded from myshell-ai/OpenVoiceV2 on first run

License

MIT License — free for commercial and research use. See LICENSE.

Credits

• Zengyi Qin @ MIT
• Wenliang Zhao @ Tsinghua
• Xumin Yu @ Tsinghua
• Ethan Sun @ MyShell