Generate, clone, design and fine-tune voices with 10 model families, including VibeVoice, Qwen3-TTS, VoxCPM2, Chatterbox, OmniVoice, Echo-TTS, MOSS-TTS and Kokoro.
OpenVoiceLab is a local web app for running open text-to-speech models. Pick a model, load it, type some text, choose a voice and click Generate. There are no notebooks or config files to deal with. When a new model comes out, you can try it next to the ones you already use.
It also covers fine-tuning: build a dataset from your recordings on the Data tab and train a VibeVoice LoRA adapter on the Training tab.
git clone https://github.com/fakerybakery/openvoicelab && cd openvoicelab
./scripts/setup.sh # scripts\setup.bat on Windows
./scripts/run.sh # scripts\run.bat on WindowsOpen http://localhost:7860 in your browser. See the installation guide for details.
| Model | Best for | License |
|---|---|---|
| VibeVoice | Long multi-speaker podcasts, fine-tuning | MIT |
| Qwen3-TTS | Preset voices with style instructions, voice design, cloning | Apache-2.0 |
| VoxCPM | 48 kHz cloning and voice design in 30 languages | Apache-2.0 |
| Chatterbox | Fast cloning, emotion control, 23 languages | MIT |
| OmniVoice | Cloning in 600+ languages | ⚠️ CC-BY-NC-4.0 |
| Echo-TTS | Expressive English cloning | ⚠️ CC-BY-NC-SA-4.0 |
| MOSS-TTS | Cloning in 31 languages, 5-speaker dialogue, voice design | Apache-2.0 |
| Kitten TTS | Tiny and fast on CPU | Apache-2.0 |
| Pocket TTS | Fast on CPU, cloning, 7 languages | CC-BY-4.0 |
| Kokoro | Tiny preset-voice model | Apache-2.0 |
⚠️ = non-commercial. See Supported Models for checkpoints, capabilities and tips.
OpenVoiceLab is in beta. Feedback and contributions are welcome on GitHub or Discord.