Robust One-step Speech Enhancement via Consistency Distillation (ROSE-CD)(IEEE WASPAA ORAL)

This repository accompanies the following paper:

Robust One-step Speech Enhancement via Consistency Distillation(IEEE WASPAA ORAL)
Liang Xu, Longfei Felix Yan, W. Bastiaan Kleijn
IEEE Workshop on Applications of Signal Processing to Audio and Acoustics (WASPAA), 2025

🔗 Project Website
📄 Read the Paper (arXiv)

---

Highlights

• A one-step consistency distillation method for real-time speech enhancement.
• Mitigates teacher bias via randomized training and time-domain auxiliary losses.
• Achieves 54× faster inference while surpassing the 30-step teacher model.
• Demonstrates strong generalization across out-of-domain and real-world scenarios.

---

Citation

If you find this work helpful, please cite:

@INPROCEEDINGS{xu2025rosecd,
  author={Xu, Liang and Yan, Longfei Felix and Kleijn, W. Bastiaan},
  booktitle={2025 IEEE Workshop on Applications of Signal Processing to Audio and Acoustics (WASPAA)}, 
  title={Robust One-step Speech Enhancement via Consistency Distillation}, 
  year={2025},
  volume={},
  number={},
  pages={1-5},
  keywords={Noise;Speech enhancement;Robustness;Real-time systems;Trajectory;Recording;Noise measurement;Iterative methods;Time-domain analysis;Optimization},
  doi={10.1109/WASPAA66052.2025.11230988}}