A demonstration application for TCSinger2 - Customizable Multilingual Zero-shot Singing Voice Synthesis, optimized for macOS with Apple Metal GPU support.

✨ Zero-shot Singing Voice Synthesis: Synthesize singing voices without training on specific singers 🎤 Timbre Control: Use audio prompts to define voice style and timbre 🎵 Musical Control: Specify lyrics and musical notes for synthesis 🖥️ macOS Optimized: Adapted for Apple Metal Performance Shaders (MPS) 🎨 User-Friendly UI: Built with Gradio for easy interaction

TCSinger 2 is a state-of-the-art multilingual zero-shot singing voice synthesis model that supports:

• Style transfer from audio prompts
• Multi-level style control
• Cross-lingual synthesis
• Speech-to-singing conversion

Paper: TCSinger 2: Customizable Multilingual Zero-shot Singing Voice Synthesis (ACL 2025)

• macOS with Apple Silicon (M1/M2/M3) or Intel Mac
• Python 3.10 or higher
• At least 8GB RAM (16GB recommended)

git clone https://github.com/audiohacking/TCsinger-App.git
cd TCsinger-App

python3 -m venv venv
source venv/bin/activate

For Apple Silicon (M1/M2/M3):

# Install PyTorch with MPS support
pip install torch torchvision torchaudio
# Install other dependencies
pip install -r requirements.txt

For Intel Macs:

pip install -r requirements.txt

python -c "import torch; print('PyTorch:', torch.__version__); print('MPS Available:', torch.backends.mps.is_available())"

Launch the demo application:

python app/demo.py

The application will start and open in your default browser at http://127.0.0.1:7860

python app/demo.py --help

Available options:

• --model-path: Path to pretrained model checkpoint (optional)
• --share: Create a public share link for remote access
• --server-name: Server host (default: 127.0.0.1)
• --server-port: Server port (default: 7860)

Example:

python app/demo.py --share --server-port 8080

1. Upload Audio Prompt:

   • Upload a short audio clip (3-10 seconds) or record using your microphone
   • This defines the timbre/voice style you want to replicate

2. Enter Lyrics:

   • Type the text you want to be sung
   • Supports multiple languages

3. Specify Musical Notes:

   • Use note names: C4 D4 E4 F4 G4
   • Or MIDI numbers: 60 62 64 65 67
   • Or mixed with rests: C4 rest E4 G4

4. Adjust Settings (Optional):

   • CFG Scale: Controls adherence to prompts (1.0-10.0)
   • Higher values = more faithful to input prompts

5. Synthesize:

   • Click "Synthesize Singing Voice"
   • Wait for processing
   • Download or play the result

TCsinger-App/
├── app/
│ ├── __init__.py
│ ├── config.py # Configuration settings
│ └── demo.py # Main Gradio application
├── utils/
│ ├── __init__.py
│ ├── audio_utils.py # Audio processing utilities
│ └── model_loader.py # Model loading and inference
├── models/ # Model checkpoints directory
├── requirements.txt # Python dependencies
└── README.md # This file

This demo is designed to work with TCSinger2 pretrained models. To use a pretrained model:

1. Download model checkpoints from TCSinger2 repository
2. Place checkpoints in the models/ directory
3. Launch with: python app/demo.py --model-path models/your_checkpoint.pt

Note: Currently, the demo runs with a placeholder model for demonstration. For full functionality, you'll need to:

• Clone and set up the TCSinger2 repository
• Train or download pretrained models
• Integrate the actual model code

This application is optimized for Apple Metal GPUs:

• Automatically detects and uses MPS (Metal Performance Shaders) when available
• Falls back to CPU if MPS is not available
• Significantly faster inference on Apple Silicon Macs

To verify Metal support:

import torch
print(torch.backends.mps.is_available()) # Should print True on Apple Silicon

python -m pytest tests/

# Format code
black app/ utils/
# Check style
flake8 app/ utils/

• Custom model training interface
• Batch processing support
• Advanced audio preprocessing options
• More language support
• Export in multiple formats
• Integration with DAWs

• Ensure you're on macOS 12.3 or later with Apple Silicon
• Update to the latest PyTorch version

• Make sure all dependencies are installed: pip install -r requirements.txt
• Activate the virtual environment: source venv/bin/activate

• Try adjusting the CFG scale
• Use a higher quality audio prompt
• Ensure lyrics and notes are properly aligned

If you use TCSinger2 in your research, please cite:

@article{zhang2025tcsinger,
 title={TCSinger 2: Customizable Multilingual Zero-shot Singing Voice Synthesis},
 author={Zhang, Yu and Guo, Wenxiang and Pan, Changhao and Yao, Dongyu and Zhu, Zhiyuan and Jiang, Ziyue and Wang, Yuhan and Jin, Tao and Zhao, Zhou},
 journal={arXiv preprint arXiv:2505.14910},
 year={2025}
}

This project is for demonstration and research purposes. Please refer to the TCSinger2 repository for licensing information.

⚠️ Important: This technology should only be used ethically and legally:

• Do not generate singing voices of public figures without permission
• Respect copyright and intellectual property rights
• Obtain consent before using someone's voice
• Follow all applicable laws and regulations

• TCSinger2 by Zhejiang University
• Gradio for the UI framework
• PyTorch for deep learning
• Apple Metal Performance Shaders for GPU acceleration

For issues related to:

• This demo app: Open an issue in this repository
• TCSinger2 model: Visit the official repository

Made with ❤️ for the audio synthesis community