chatterbox-turbo-apple-silicon

Chatterbox TTS with MLX on Apple Silicon 🎙️

High-quality voice synthesis with emotion control and voice cloning using Chatterbox Turbo on Apple Silicon.

Features

• 🚀 Fast inference on Apple Silicon (M1/M2/M3/M4) using MLX
• 🎭 9 Emotion Tags - [laugh], [sigh], [gasp], [groan], [chuckle], [cough], [sniff], [shush], [clear throat]
• 🎵 Voice Cloning from reference audio (6+ seconds)
• 📝 Smart Text Chunking for long inputs
• 🎨 Professional Web UI with Gradio

Quick Start

# 1. Install dependencies
pip install -r requirements.txt

# 2. Launch the web interface
./start_demo.sh

# Or run directly:
python3 app_chatterbox.py

# 3. Open in browser (auto-opens)
# URL: http://127.0.0.1:7861

If Already Running

# Check if running
pgrep -f app_chatterbox.py

# Stop the demo
pkill -f app_chatterbox.py

# Then start again
./start_demo.sh

Using the Web Interface

1. Enter text with optional emotion tags like [laugh] or [sigh]
2. Upload reference audio (optional) for voice cloning
3. Click Generate to create speech
4. Listen to the generated audio with emotions!

Command Line Usage

# Basic generation with emotions
python3 -m mlx_audio.tts.generate \
  --model mlx-community/chatterbox-turbo-fp16 \
  --text "[sigh] Monday again. [chuckle] But let's make the best of it!" \
  --file_prefix output

# Voice cloning
python3 -m mlx_audio.tts.generate \
  --model mlx-community/chatterbox-turbo-fp16 \
  --text "Hello, this is your cloned voice!" \
  --ref_audio my_voice.wav \
  --file_prefix cloned

Supported Emotion Tags

Example Audio

Check out chatterbox_full_story.wav - an 82-second story with multiple emotions!

Voice Cloning Tips

1. Use clear audio - Record in a quiet environment
2. 6+ seconds - Longer samples clone better
3. Single speaker - Only one person speaking
4. Good quality - WAV format recommended

Technical Details

• Model: mlx-community/chatterbox-turbo-fp16 (350M parameters)
• Framework: MLX (Apple's machine learning framework)
• Sample Rate: 24kHz
• Pipeline: Chatterbox Turbo (not Kokoro)

Files

Installation

# Core packages
pip install mlx-audio[tts]>=0.2.8
pip install librosa gradio

# Or use requirements.txt
pip install -r requirements.txt

Requirements

• Hardware: Apple Silicon Mac (M1/M2/M3/M4)
• Python: 3.10+
• mlx-audio: 0.2.8+
• Dependencies: librosa, gradio, soundfile

Why This Works

The key insight: Use CLI module (python -m mlx_audio.tts.generate) instead of the Python API's generate_audio() function. The CLI properly routes to Chatterbox Turbo pipeline, while the Python function defaults to Kokoro.

You can verify it's using Chatterbox when you see:

S3 Token -> Mel Inference...

(Not "KokoroPipeline")

Credits

• Chatterbox Model: Resemble AI
• MLX Audio: Prince Canuma / Blaizzy
• MLX Framework: Apple

License

MIT License - Free to use and modify!

---

🎉 Enjoy natural, expressive speech synthesis with emotion control on your Mac!