DSRN ICASSP 2023

SPEECH AND NOISE DUAL-STREAM SPECTROGRAM REFINE NETWORK WITH SPEECH DISTORTION LOSS FOR ROBUST SPEECH RECOGNITION

Data Link

The simulated data link is: DSRNet-data.

The noise link is: noise-data.

There may be a delay. Please wait one minute.

arxiv Paper Link

There's been a slight error in the equation (9) in our paper, and the link to the latest version of the paper is: new-paper-link-arxiv.

Code

Our code is based on espnet. You can see it in espnet-dsrn/egs2/aishell_noise.

Run

You can run this code with

cd ./espnet-dsrn/egs2/aishell_noise
bash run_dsrn_fbank.sh

Citation

We kindly request that our work be cited in relevant academic discussions. Please refer to the citation details provided below.

@inproceedings{lu2023speech,
  title={speech and noise dual-stream spectrogram refine network with speech distortion loss for robust speech recognition},
  author={Lu, Haoyu and Li, Nan and Song, Tongtong and Wang, Longbiao and Dang, Jianwu and Wang, Xiaobao and Zhang, Shiliang},
  booktitle={ICASSP 2023-2023 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)},
  pages={1--5},
  year={2023},
  organization={IEEE}
}

Contact

For technical support, dataset inquiries, or collaboration opportunities, please contact us: [email protected]