An Improved Real-Time Noise Suppression Method Based on RNN and Long-Term Speech Information

Baoping Cheng, Baoping Cheng, Guisheng Zhang, Xiaoming Tao, Sheng Wang, Nan Wu, Min Chen

2022

Abstract

Speech enhancement based on deep learning can provide almost best performance when processing non-stationary noise. Denoising methods that combine classic signal processing with Recurrent Neural Network (RNN) can be implemented in real-time applications due to their low complexity. However, long term speech information is omitted when selecting the features in these methods, which degrades the denoising performance. In this paper, we extend a well-known RNN based denoising method called RNNoise with the long-term spectral divergence (LTSD) feature. We also limited the amount of noisy speech attenuation to get a better trade-off between noise removal level and speech distortion. Our proposed method outperforms the RNNoise algorithm by 0.12 MOS points on average in the subjective listening test.

Paper Citation

in Harvard Style

Cheng B., Zhang G., Tao X., Wang S., Wu N. and Chen M. (2022). An Improved Real-Time Noise Suppression Method Based on RNN and Long-Term Speech Information. In Proceedings of the 3rd International Symposium on Automation, Information and Computing - Volume 1: ISAIC; ISBN 978-989-758-622-4, SciTePress, pages 795-800. DOI: 10.5220/0012055400003612

in Bibtex Style

@conference{isaic22,
author={Baoping Cheng and Guisheng Zhang and Xiaoming Tao and Sheng Wang and Nan Wu and Min Chen},
title={An Improved Real-Time Noise Suppression Method Based on RNN and Long-Term Speech Information},
booktitle={Proceedings of the 3rd International Symposium on Automation, Information and Computing - Volume 1: ISAIC},
year={2022},
pages={795-800},
publisher={SciTePress},
organization={INSTICC},
doi={10.5220/0012055400003612},
isbn={978-989-758-622-4},
}

in EndNote Style

TY - CONF

JO - Proceedings of the 3rd International Symposium on Automation, Information and Computing - Volume 1: ISAIC
TI - An Improved Real-Time Noise Suppression Method Based on RNN and Long-Term Speech Information
SN - 978-989-758-622-4
AU - Cheng B.
AU - Zhang G.
AU - Tao X.
AU - Wang S.
AU - Wu N.
AU - Chen M.
PY - 2022
SP - 795
EP - 800
DO - 10.5220/0012055400003612
PB - SciTePress