Chatterbox TTS — Zee's Voice

Self-hosted TTS API using Chatterbox (MIT license, beats ElevenLabs in blind tests at 63.75% preference).

Quick Start (Google Colab — Free T4 GPU)

1. Open chatterbox-colab.ipynb in Google Colab
2. Runtime → Change runtime type → T4 GPU
3. Run Cell 1 (install deps)
4. Run Cell 2 (start server + ngrok tunnel)
5. Copy the ngrok URL, paste in Discord for Zee

Files

• server.py — FastAPI TTS server (OpenAI-compatible endpoints)
• client.py — VPS-side client to call the server
• chatterbox-colab.ipynb — One-click Colab notebook

API Endpoints

Endpoint	Method	Description
/health	GET	Health check + GPU info
/voices	GET	List available voices
/v1/audio/speech	POST	OpenAI-compatible TTS
/speak	POST	Extended control (exaggeration, cfg_weight)

Usage from VPS

# Set the server URL (from Colab ngrok output)
export CHATTERBOX_URL="https://xxxx.ngrok.io"

# Generate speech
python /root/chatterbox-tts/client.py "Hey Darko, it's Zee."

# Check health
python /root/chatterbox-tts/client.py --health

Requirements (Server-side)

• Python 3.10+
• CUDA GPU with 4GB+ VRAM (T4 works for turbo variant)
• Packages: chatterbox-tts, fastapi, uvicorn, torch, torchaudio

Model

• Chatterbox-Turbo: 350M params, 4-8GB VRAM, ~6x realtime on T4
• Chatterbox-Full: 500M params, 8-16GB VRAM, best emotion control
• License: MIT
• Source: https://github.com/resembleai/chatterbox