The performance of speech communication and automatic speech recognition systems greatly degrades in real conditions where reverberation and noise exist. Traditional noise reduction algorithms or dereverberation algorithms could suppress the corresponding interference effectively. But when noise and reverberation exist at the same time, the performance of various speech enhancement algorithms degrades significantly.
To jointly deal with the effect of reverberation and noise, researchers from the Institute of Acoustics of the Chinese Academy of Sciences (IACAS) proposed an integrated multi-channel speech enhancement approach for joint noise reduction and dereverberation based on the multi-channel Wiener filter (MWF) and the weighted prediction error (WPE).
In the proposed method, researchers decomposed the MWF into a MVDR (minimum variance distortionless response) beamformer followed by a single-channel Wiener filter. Instead of the entire relative transfer functions (RTFs) or the direct-path of desired speech signal, researchers used the relative early transfer functions (RETFs) in the MVDR beamformer. They approximated the covariance matrix of the undesired signal using the prediction provided by the WPE method. The target signal variance in WPE was further updated with the output signal.
Researchers conducted experiments in the simulated and real conditions and results showed that the proposed algorithm could enhance the speech signal effectively where noise and reverberation existed simultaneously. The word error rate was reduced by 38.50% in 3 m recorded data conditions. By comparing with other five state-of-the-art algorithms, the proposed algorithm outperformed in terms of speech quality, speech intelligibility and speech recognition performance.
The research results show that the proposed algorithm could improve the user experience of voice communication systems and automatic voice recognition systems significantly when both noise and reverberation exist.
The research, published online in Applied Acoustics, was supported by the National Key Research and Development Program and the National Natural Science Foundation of China.
Figure 1. The block diagram of the proposed joint noise reduction and dereverberation approach. (Image by IACAS)
Reference:
SONG Siyuan, CHENG Longbiao, LUAN Shuming, YAO Dingding, LI Junfeng, YAN Yonghong. An integrated multi-channel approach for joint noise reduction and dereverberation. Applied Acoustics, 171: 107526. DOI: 10.1016/j.apacoust.2020.107526.
Contact:
ZHOU Wenjia
Institute of Acoustics, Chinese Academy of Sciences, 100190 Beijing, China
E-mail: media@mail.ioa.ac.cn