HQ-SVC: Towards High-Quality Zero-Shot Singing Voice Conversion in Low-Resource Scenarios

Official Repository of Paper: "Towards High-Quality Zero-Shot Singing Voice Conversion in Low-Resource Scenarios"(AAAI 2026)

HQ-SVC Logo

🗞 News

[2025-11-08] 🎉 Paper accepted by AAAI 2026
[2025-11-12] 🎉 arXiv paper released
[2025-11-12] 🎉 Demo released
[2025-12-24] 🎉 Inference codes and pre-trained models released

📅 Release Plan

arXiv preprint
Online demo
Inference codes
Pre-trained models
Training codes

✨ New features

Singing style control
Improved quality

HQ-SVC is an efficient framework for high-quality zero-shot singing voice conversion (SVC) in low-resource scenarios. It achieves disentanglement of content and speaker features via a unified decoupled codec, and enhances synthesis quality through multi-feature fusion and progressive optimization.

Unlike existing methods that demand large datasets or heavy computational resources, HQ-SVC unifies:

🚀 Zero-shot conversion for unseen speakers without fine-tuning
⚡ Low-resource training (single consumer-grade GPU, <80h data)
🎧 Dual capabilities: high-quality singing voice conversion + voice super-resolution
🎯 Superior naturalness and speaker similarity compared to SOTA methods

🎸 Try Inference

1. Download Codes and Environment 下载代码和环境

git clone https://github.com/ShawnPi233/HQ-SVC.git
cd HQ-SVC

wget -c https://huggingface.co/shawnpi/HQ-SVC/resolve/main/environment.tar.gz

wget -c https://hf-mirror.com/shawnpi/HQ-SVC/resolve/main/environment.tar.gz # 可选镜像源

2. Unzip Environment 解压环境

mkdir -p venv
tar -xzf environment.tar.gz -C venv

3. Activate Environment 激活环境

source venv/bin/activate

4. Download Pretrained Models 下载权重

export HF_HUB_ENABLE_HF_TRANSFER=0
huggingface-cli download shawnpi/HQ-SVC --include "utils/pretrain/*" --local-dir . --local-dir-use-symlinks False

5. Running 运行

python gradio_app.py

如果报错 Caught signal 11 (Segmentation fault: address not mapped to object at address (nil))
请执行以下代码后再启动上述代码

unset LD_LIBRARY_PATH

Zero-shot Super-Resolution (16 kHz to 44.1 kHz): Input only source audio

Zero-shot Singing Voice Conversion: Input both source audio and target audio

📜 Citation

If you use HQ-SVC in your research, please cite our work:

@article{bai2025hq,
  title={HQ-SVC: Towards High-Quality Zero-Shot Singing Voice Conversion in Low-Resource Scenarios},
  author={Bai, Bingsong and Geng, Yizhong and Wang, Fengping and Wang, Cong and Guo, Puyuan and Gao, Yingming and Li, Ya},
  journal={arXiv preprint arXiv:2511.08496},
  year={2025}
}

🙏 Acknowledgement

We thank the open-source communities behind:

⭐️ Star History

Downloads last month: -; Downloads are not tracked for this model. How to track