🎙️ Chatterbox TTS Proxy

Ever wanted to use ChatGPT's voice features with your own custom voices? This proxy makes it happen! It's a simple bridge that lets any app using OpenAI's TTS API tap into Chatterbox's awesome voice cloning powers. No code changes needed - just point your app at this proxy and you're good to go! 🚀

✨ Why Use This Proxy?

• Drop-in Replacement - Works with any OpenAI TTS client - just swap the endpoint URL
• Zero-shot Voice Cloning - Clone any voice from a single audio sample
• Emotional Control - Make your AI sound happy, sad, or anywhere in between
• Production Ready - High-quality audio that's ready for real-world use
• Easy Deployment - One command to get everything running

🎯 Chatterbox TTS

Chatterbox is Resemble AI's open-source TTS engine that's seriously impressive:

• Clone any voice from a single audio sample (yes, really)
• Control how emotional the voice sounds
• Generate studio-quality audio
• Works with WAV, MP3, and FLAC files

This project uses bhimrazy's Chatterbox TTS server from Docker Hub. It's the perfect starting point for voice cloning experiments!

🔄 Adapter Proxy

The adapter translates between OpenAI's TTS API format and Chatterbox's more powerful capabilities:

• Converts OpenAI requests to Chatterbox format
• Manages voice presets and settings
• Handles errors gracefully
• Makes everything just work™

Here's how it works:

Your app sends this (standard OpenAI format):

{
  "model": "tts-1", #this value is ignored
  "voice": "en-uk-heart",
  "input": "Hello, this is a test."
}

The proxy converts it to this (Chatterbox format):

{
  "text": "Hello, this is a test.",
  "audio_prompt": "reference/voice_sample.wav",
  "cfg": 0.4,          # Controls generation quality
  "exaggeration": 0.6,  # Controls emotional intensity
  "temperature": None, # Only needed when an audio prompt isn't provided
}

The magic happens with these settings:

• audio_prompt - Your reference audio file (WAV, FLAC, or MP3)
• exaggeration - How emotional the voice should be (0.0-1.0)
• cfg - How closely to match the reference voice (0.0-1.0)
• temperature - If you have no reference file, this controls the randomness of the voice

🎭 Voice Cloning

Want to add your own voice? It's super easy! Here's how:

1. Drop your audio file in the reference/ folder (5-45 seconds will work, but I recommend at least 15-20 seconds for accuracy)
2. Add a new preset to adapter.py
3. Tweak the settings:
   • exaggeration: Crank it up (0.8-1.0) for more dramatic speech
   • cfg: Higher values (0.7-0.9) for better voice matching

That's it! The proxy handles all the complicated stuff behind the scenes. 🎯

🚀 Setup & Run

Docker Compose

Fire it up with one command:

docker-compose up -d

📁 Folder Structure

.
├── models/       # Hugging Face models (cached)
├── reference/    # Your voice samples go here
├── voices.yml    # Voice preset configurations
└── docker-compose.yml

🎭 Voice Configuration

The voices.yml file lets you configure your voice presets. Each preset can use either:

• An audio sample for voice cloning (recommended)
• Temperature-based generation (when no audio sample is available)

Example configuration:

# Voice with audio sample
stewie:
  name: Stewie
  audio_prompt: stewie.mp3
  exaggeration: 0.8  # How emotional (0.0-1.0)
  cfg: 0.4          # Voice matching quality (0.0-1.0)

# Voice without audio sample
random-voice:
  name: Random Voice
  temperature: 0.7  # Voice randomness (0.0-1.0)
  exaggeration: 0.5
  cfg: 0.5

To use your own voice configuration:

1. Create a voices.yml file in your project directory
2. Mount it in the container using the volume in docker-compose.yml
3. Optionally override the path using the VOICES_FILE environment variable

The proxy will automatically reload the voice presets when the container starts.

🧪 Quick Test

You can test the proxy directly using curl:

curl -X POST http://your-server:8004/v1/audio/speech \
  -H "Content-Type: application/json" \
  -d '{
    "model": "tts-1",
    "voice": "stewie",
    "input": "How does it feel to be the least cultured person at a bus station?"
  }' \
  --output test.wav

This will:

1. Send a request to the proxy
2. Use the "stewie" voice preset
3. Save the generated audio as test.wav

You can then play the file to hear the result. Try different voices and text to experiment!

⚖️ Ethical Considerations

We strongly encourage responsible use of this technology:

• Use your own voice or public domain voices for personal projects
• Get explicit permission before cloning someone else's voice
• Be transparent when using AI-generated voices
• Consider the impact on voice actors and content creators

While we can't control how others use this tool, we don't condone:

• Impersonating others without consent
• Creating misleading or harmful content
• Using cloned voices for fraud or deception
• Any other malicious or unethical purposes

Remember: With great power comes great responsibility! 🕷️

📄 License

This project is licensed under the MIT License - see the LICENSE file for details.