🎙️ RunPod Chatterbox TTS

A serverless text-to-speech endpoint powered by Chatterbox Multilingual on RunPod. Generate natural speech in 23 languages with zero-shot voice cloning.

✨ Features

• 23 Languages — Spanish, English, French, German, Japanese, Chinese, and more
• Voice Cloning — Clone any voice with just 5-10 seconds of reference audio
• Emotion Control — Adjust expressiveness from monotone to dramatic
• Serverless — Pay only for what you use, auto-scales to zero
• Fast Cold Starts — Model pre-baked into Docker image + FlashBoot ready

🚀 Quick Start

Deploy to RunPod

1. Pull the image or build your own:

   docker build -t yourusername/chatterbox-tts:latest .
   docker push yourusername/chatterbox-tts:latest

2. Create a serverless endpoint in RunPod Console:

   • Template → New Template → Enter your Docker image
   • GPU: RTX 4000 Ada / L4 / A4000 (8-16GB VRAM)
   • Enable FlashBoot ✅

API Usage

import runpod
import base64

runpod.api_key = "your_api_key"
endpoint = runpod.Endpoint("your_endpoint_id")

# Basic TTS
result = endpoint.run_sync({
    "input": {
        "text": "Hola, esto es una prueba.",
        "language_id": "es"
    }
})

# With voice cloning
with open("reference.wav", "rb") as f:
    ref_audio = base64.b64encode(f.read()).decode()

result = endpoint.run_sync({
    "input": {
        "text": "Your text here",
        "language_id": "en",
        "reference_audio": ref_audio,
        "exaggeration": 0.6
    }
})

# Save output
audio = base64.b64decode(result["audio_base64"])
with open("output.wav", "wb") as f:
    f.write(audio)

📥 Input Schema

Field	Type	Required	Description
text	string	✅	Text to synthesize
language_id	string	✅	Language code (see below)
reference_audio	string	❌	Base64 WAV for voice cloning
exaggeration	float	❌	Emotion intensity (0.0-1.0, default 0.5)
cfg_weight	float	❌	Style adherence (0.0-1.0, default 0.5)

📤 Output Schema

{
  "audio_base64": "UklGRi...",
  "sample_rate": 24000,
  "duration_seconds": 2.45
}

🌍 Supported Languages

Code	Language	Code	Language	Code	Language
ar	Arabic	he	Hebrew	pl	Polish
da	Danish	hi	Hindi	pt	Portuguese
de	German	it	Italian	ru	Russian
el	Greek	ja	Japanese	sv	Swedish
en	English	ko	Korean	sw	Swahili
es	Spanish	ms	Malay	tr	Turkish
fi	Finnish	nl	Dutch	zh	Chinese
fr	French	no	Norwegian

🎯 Voice Cloning Tips

For best results:

• Use 5-15 seconds of clean audio
• WAV format, 24kHz+ sample rate
• Single speaker, no background noise
• Match the reference style to desired output emotion

💰 Cost Estimation

Traffic	GPU	Active Workers	~Monthly Cost
100 req/day	RTX 4000 Ada	0 (flex)	$5-15
1,000 req/day	L4	1	$50-80
10,000+ req/day	L4	2+	$200+

🛠️ Local Development

# Test locally
python handler.py --test_input '{"input": {"text": "Hello world", "language_id": "en"}}'

📄 License

MIT — Model weights subject to Chatterbox license.

🙏 Credits

• Resemble AI — Chatterbox TTS model
• RunPod — Serverless GPU infrastructure