Research On Multi-Channel Speech Enhancement Method In Vehicle Environment

In the vehicle environment,the performance of traditional multi-channel speech enhancement methods is poor.It is of great significance to study the speech enhancement technology in this environment.A multi-channel feature extraction method based on attention mechanism is proposed to solve the problem of poor effect of traditional methods in extracting features between channels.Firstly,this method uses the attention mechanism to assign different weight values to the frame level data of each channel to increase the difference between channels,so that the model can better learn the spatial characteristics between channels,and then further analyze the impact of different microphone structures on the experimental results.The experimental results show that compared with NCC and ICD feature extraction methods,the proposed method improves the evaluation indexes STOI,PESQ,SI-SNR and SDR by 2.2%,1.5%,1.8% and 1.6% on average.In view of the fact that using a single method to extract channel features can not fully represent the spatial information of microphone array,a multi-channel speech enhancement method based on multi-feature fusion is proposed.Firstly,the method uses the multi-feature fusion strategy to fuse the features extracted by different channel feature extraction methods,and then uses the two-stage structure to extract the channel features for many times,making full use of the features between channels.The experimental results show that the channel feature extraction method after feature fusion can achieve better noise reduction effect.Compared with NCC and ICAM feature extraction methods,the proposed method has improved the evaluation indexes STOI,PESQ,SI-SNR and SDR by 3.3%,1.8%,2% and 2.1% on average.This paper has 31 figures,8 tables and 50 references.