Skip to content

Nested U-Net with two-level skip connections for speech enhancement

License

Notifications You must be signed in to change notification settings

seorim0/NUNet-TLS

Repository files navigation

Monoaural Speech Enhancement Using a Nested U-Net with Two-Level Skip Connections

This is a repo of the paper "Monoaural Speech Enhancement Using a Nested U-Net with Two-Level Skip Connections", which is accepted to INTERSPEECH2022.

Abstract:Capturing the contextual information in multi-scale is known to be beneficial for improving the performance of DNN-based speech enhancement (SE) models. This paper proposes a new SE model, called NUNet-TLS, having two-level skip connections between the residual U-Blocks nested in each layer of a large U-Net structure. The proposed model also has a causal time-frequency attention (CTFA) at the output of the residual U-Block to boost dynamic representation of the speech context in multi-scale. Even having the two-level skip connections, the proposed model slightly increases the network parameters, but the performance improvement is significant. Experimental results show that the proposed NUNet-TLS has superior performance in various objective evaluation metrics to other state-of-the-art models.

Requirements

This repo is tested on Ubuntu 20.04.

# for train
python == 3.7.9   
pytorch == 1.9.0_cu111   
scipy == 1.6.0      
soundfile == 0.10.3  
# for evaluation
tensorboard == 2.7.0   
pesq == 0.0.2       
pystoi == 0.3.3       
matplotlib == 3.3.3      

Getting started

  1. Install the necessary libraries.
  2. Set directory paths for your dataset. (config.py)
# dataset path
noisy_dirs_for_train = '../Dataset/train/noisy/'   
clean_dirs_for_train = '../Dataset/train/clean/'   
noisy_dirs_for_valid = '../Dataset/valid/noisy/'   
clean_dirs_for_valid = '../Dataset/valid/clean/'   
  • You need to modify the find_pair function in tools.py according to the data file name you have.
  • And if you need to adjust any parameter settings, you can simply change them.
  1. Run train_interface.py

Results

Demo

We randomly select one sample for demonstration at 10 dB SNR.

1_Clean.mov
1_Noisy.mov
1_DCCRN+C.mov
1_FullSubNet.mov
1_SADNUNet.mov
1_NUNet-TLS.mov

References

U2-Net: Going deeper with nested u-structure for salient object detection
X. Qin, Z. Zhang, C. Huang, M. Dehghan, O. R. Zaiane, and M. Jagersand
[paper] [code]
A nested u-net with self-attention and dense connectivity for monaural speech enhancement
X. Xiang, X. Zhang, and H. Chen
[paper]
Time-frequency attention for monaural speech enhancement
Q. Zhang, Q. Song, Z. Ni, A. Nicolson, and H. Li
[paper]

Releases

No releases published

Packages

No packages published

Languages