Research Of Microphone Array Speech Signal Denoising Methods Based On Tensor Analysis

Because of the interference of background noise, speech signal received from the microphone array is usually accompanied with noise in voice communication systems, such as phone call, video conference and hand-free car system. Such noisy speech will own some intelligible and comfortable problems or influence the processing performance of subsequent processing system. Traditional single microphone speech denoising methods such as wiener filtering is widely used because of its simplicity and efficiency. However,such single channel algorithm may cause signal distortion or introduce the so-called musical noise. It can not meet the needs of multimedia development. So people begin to study and use microphone array speech denoising technology. The microphone array can obtain signal with time and space information simultaneously when receiving signals, so it can be more flexible to improve the performance of noise reduction. Meanwhile, the target speech source can be aimed to track the speaker’s position. The microphone array speech denoising technology can be used more widely.On the other hand, Tensor analysis has been widely studied and used in many fields such as quantum physics, text mining, data analysis and image processing with the development of multi-channel or multi-dimensional data in recent years. Tensor can be seen as a multidimensional matrix, which is a high dimensional expansion of vector and matrix.The development of tensor analysis allows people to consider modeling the signal into a tensor form so as to solve signal problems in high dimensional space. The analysis based on tensor algebra is especially suitable for analysis and processing of multidimensional array signals.The paper introduce theoretic tensor basics, general signal model of microphone arrays,time delay estimation and multi-microphone speech signal denoising methods. Some existing methods for speech denoising based on microphone array such as delay and sum beamforming algorithm, adaptive beamforming based on GSC theoretical Prototype and subspace algorithm are introduced firstly. Simulation and comparison of these methods are carried out. Then, we model the multi-microphone speech signal into a tensor form according to three dimensions of channel, time and frequency based on the above tensor algebra analysis technology. We build multi-mode linear filters to reduce noise according to Tucker decomposition and alternating least square algorithm and research methods of combing tensor pre-processing and microphone array subspace algorithm. On the whole,the methods can perform well in our experimental simulation. The research of our paper can provide some useful results to get a better noise reduction in high dimension in the future.