Chatterbox Multilingual TTS

Turn text into natural speech in 23 languages. Clone voices too.

Built by ResembleAI, optimized for fast deployment.

What this does

This model takes your text and turns it into realistic speech. Works in 23 languages: Arabic, Chinese, Danish, Dutch, English, Finnish, French, German, Greek, Hebrew, Hindi, Italian, Japanese, Korean, Malay, Norwegian, Polish, Portuguese, Russian, Spanish, Swedish, Swahili, Turkish.

Upload your own audio file and it'll try to clone that voice. Pretty cool.

Why it's cool

• Fast: Downloads 3GB of model weights in 17 seconds (thanks to pget)
• Multilingual: Actually works well across 23 languages
• Voice cloning: Upload a voice sample, get speech in that voice
• Just works: No HuggingFace auth, no complicated setup

How to run it

git clone https://github.com/zsxkib/cog-ResembleAI-Chatterbox-Multilingual-TTS
cd cog-ResembleAI-Chatterbox-Multilingual-TTS
cog predict -i text="Hello world" -i language="en"

That's it. Cog will build the container, download the weights, and generate speech.

Want to clone a voice? Add a reference audio file:

cog predict -i text="Hello in my voice" -i language="en" -i reference_audio=@voice.wav

Want more drama in the speech?

cog predict -i text="Very dramatic speech" -i language="en" -i exaggeration=0.8

Requirements

• Cog
• Docker
• That's it

The first run downloads model weights automatically. Subsequent runs are fast.

License

MIT