UL-UNAS

🎉 This is the official implementation of our IEEE TASLP paper:

UL-UNAS: Ultra-Lightweight U-Nets for Real-Time Speech Enhancement via Network Architecture Search

🔥 News

• [2026-2-13] Added streaming inference code ulunas_onnx implemented by Kailai Shen.
• [2026-2-3] The updated paper is uploaded to arxiv.
• [2026-2-1] The pre-trained checkpoint is released.
• [2026-1-28] The model implementation is released.

Inference

To run inference on audio files, use:

python inference --input_dir <input_dir> --output_dir <output_dir> [options]

Argument	Requirement / Default	Description
--input_dir	required	Path to the input directory containing audio files.
--output_dir	required	Path to the output directory where enhanced files will be saved.
--device	default: cuda:0	Torch device to run inference on, e.g., cuda:0, cuda:1, or cpu.
--extension	default: .wav	Audio file extension to process.

Citation

If you find this work useful, please cite our paper:

@misc{rong2025ulunas,
      title={UL-UNAS: Ultra-Lightweight U-Nets for Real-Time Speech Enhancement via Network Architecture Search}, 
      author={Xiaobin Rong and Dahan Wang and Yuxiang Hu and Changbao Zhu and Kai Chen and Jing Lu},
      year={2025},
      eprint={2503.00340},
      archivePrefix={arXiv},
      primaryClass={eess.AS},
      url={https://arxiv.org/abs/2503.00340}, 
}

Contact

Xiaobin Rong: xiaobin.rong@smail.nju.edu.cn