This is a pre-trained version of Fast FullSubNet, a real-time denoising model trained on the Deep Noise Suppression Challenge dataset of 2020 (DNS-INTERSPEECH-2020).
How to run
https://fullsubnet.readthedocs.io/en/latest/usage/getting_started.html
Code
https://github.com/Audio-WestlakeU/FullSubNet
Note: The code doesn't support real-time streaming out of the box. See issue-67 for details.
Paper
Fast FullSubNet: Accelerate Full-band and Sub-band Fusion Model for Single-channel Speech Enhancement, Xiang Hao, Xiaofei Li
For many speech enhancement applications, a key feature is that system runs on a real-time, latency-sensitive, battery-powered platform, which strictly limits the algorithm latency and computational complexity. In this work, we propose a new architecture named Fast FullSubNet dedicated to accelerating the computation of FullSubNet. Specifically, Fast FullSubNet processes sub-band speech spectra in the mel-frequency domain by using cascaded linear-to-mel full-band, sub-band, and mel-to-linear full-band models such that frequencies involved in the sub-band computation are vastly reduced. After that, a down-sampling operation is proposed for the sub-band input sequence to further reduce the computational complexity along the time axis. Experimental results show that, compared to FullSubNet, Fast FullSubNet has only 13% computational complexity and 16% processing time, and achieves comparable or even better performance.
Performance
| With Reverb | No Reverb | ||||||
|---|---|---|---|---|---|---|---|
| Method | WB-PESQ | NB-PESQ | SI-SDR | STOI | WB-PESQ | NB-PESQ | SI-SDR |
| Fast FullSubNet (118 Epochs) | 2.882 | 3.42 | 15.33 | 0.9233 | 2.694 | 3.222 | 16.34 |
| FullSubNet (58 Epochs) (just for comparison) | 2.987 | 3.496 | 15.756 | 0.926 | 2.889 | 3.385 | 17.635 |