Chatterbox TTS Wellness Worker

RunPod Serverless worker for Chatterbox Text-to-Speech with voice cloning.

Capabilities

• Text-to-Speech - Natural speech synthesis
• Voice Cloning - Clone any voice from a reference audio sample
• Emotion Tags - Control emotion in speech
• Speed Control - Adjust playback speed

Deployment on RunPod

1. Fork this repo to your GitHub

2. Create Serverless Endpoint

1. Go to Serverless → New Endpoint
2. Source: GitHub repo URL
3. GPU: RTX 4090 or similar (8GB+ VRAM sufficient)
4. Max Workers: 1-2
5. Idle Timeout: 5 seconds
6. No network volume needed (model is ~3GB, included in Docker image)

3. Environment Variables

None required - model is bundled.

API Usage

Basic TTS (Default Voice)

{
  "input": {
    "text": "Welcome to your daily wellness moment. Take a deep breath."
  }
}

Voice Cloning

{
  "input": {
    "text": "Hello, this is my cloned voice speaking.",
    "reference_audio_url": "https://example.com/my-voice-sample.wav"
  }
}

Or with base64 audio:

{
  "input": {
    "text": "Hello, this is my cloned voice speaking.",
    "reference_audio_base64": "UklGRi..."
  }
}

With Emotion Tags

{
  "input": {
    "text": "[happy] I'm so excited to share these wellness tips with you!"
  }
}

Supported emotions: happy, sad, angry, surprised, neutral, calm

With Parameters

{
  "input": {
    "text": "This is a calm meditation guide.",
    "temperature": 0.5,
    "exaggeration": 0.8,
    "speed": 0.9,
    "cfg_weight": 0.5
  }
}

Parameter	Default	Range	Description
temperature	0.7	0.0-1.0	Creativity/variability
exaggeration	1.0	0.5-2.0	Emotion intensity
speed	1.0	0.5-2.0	Playback speed
cfg_weight	0.5	0.0-1.0	Classifier-free guidance

Response Format

{
  "audio_base64": "UklGRi...",
  "sample_rate": 24000,
  "duration_seconds": 3.5,
  "text": "The synthesized text",
  "emotion": "happy"
}

Cost Estimates

Task	GPU	Time	Cost
TTS (30 words)	RTX 4090	~10s	~$0.001
TTS (100 words)	RTX 4090	~30s	~$0.003
Voice Clone Setup	RTX 4090	~5s	~$0.0005

Reference Audio Requirements

For best voice cloning results:

• WAV format (16-bit, mono)
• 5-15 seconds of clear speech
• Minimal background noise
• Single speaker only
• Natural speaking pace

Local Development

# Build
docker build -t chatterbox-wellness-worker .

# Run (requires NVIDIA GPU)
docker run --gpus all chatterbox-wellness-worker

Example: Wellness Narration

{
  "input": {
    "text": "[calm] Welcome to today's mindfulness session. Find a comfortable position, close your eyes, and let's begin our journey to inner peace.",
    "temperature": 0.5,
    "speed": 0.85
  }
}

License

MIT