FullSubNet

Full-Band and Sub-Band Fusion Model for Real-Time Single-Channel Speech Enhancement

The documentation is hosted on Read the Docs. Check the documentation for how to train and test models.

• Improved FullSubNet: Further reduces computational costs and enables high sampling rate data processing, e.g., 48 KHz and 24 KHz.
  • ❇️ Model Architecture
• 📰 FullSubNet: A Full-Band and Sub-Band Fusion Model for Real-Time Single-Channel Speech Enhancement, ICASSP 2021
  • 📸 Demo (Audio Clips)
  • 🎏 Model Checkpoints
  • ❇️ Model Architecture
• 📰 Fast FullSubNet: Accelerate Full-band and Sub-band Fusion Model for Single-channel Speech Enhancement
  • ❇️ Model Architecture
  • 📸 Demo (Audio Clips)
• cIRM-based Fullband baseline model (described in the original FullSubNet paper)
  • ❇️ Model Architecture

If you use this code for your research, please consider citeing:

@INPROCEEDINGS{hao2020fullsubnet,
    author={Hao, Xiang and Su, Xiangdong and Horaud, Radu and Li, Xiaofei},
    booktitle={ICASSP 2021 - 2021 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)},
    title={Fullsubnet: A Full-Band and Sub-Band Fusion Model for Real-Time Single-Channel Speech Enhancement},
    year={2021},
    pages={6633-6637},
    doi={10.1109/ICASSP39728.2021.9414177}
}

This repository Under the MIT license.