Multi-channel voice signal enhancement method applied to microphone array

Abstract

The invention discloses a multi-channel voice signal enhancement method applied to a microphone array. The multi-channel voice signal enhancement method comprises the following steps: generating training data; preprocessing the training data; performing inter-channel feature extraction on a preprocessing result; designing a multi-channel filter; performing filtering operation on the first framing voice data or the second framing voice data or the noisy voice data of the corresponding channel by using the filter of each channel to obtain a filtered enhanced voice signal; and performing weighted summation and averaging operation on the enhanced voice signals of each channel to obtain target output voice. According to the multi-channel voice signal enhancement method provided by the invention, the spatial characteristics of the noise reduction space are fully considered, the characteristics among multiple channels are extracted by adopting an innovative method, and the dual-stage Bi-LSTM is designed as a multi-channel filter, so that the sound field transmission characteristics are expressed more accurately.

Description

technical field [0001] The invention relates to the field of speech noise reduction, in particular to a multi-channel speech signal enhancement method applied to a microphone array. Background technique [0002] Traditional multi-channel speech enhancement methods are mainly based on centralized microphone arrays, and there are few studies on multi-channel speech enhancement methods for distributed microphone arrays. Algorithms for centralized microphone arrays are mostly based on signal processing methods, such as common beamforming algorithms. According to algorithm parameters, beamforming methods can be divided into fixed beamforming methods and adaptive beamforming methods. The fixed beamforming method is based on a fixed-structure microphone array. The algorithm sets the filter coefficients in each direction of the microphone array, so that the array can enhance the sound source in a specific direction. The directivity is fixed, and the common one is multi-channel. De...

AI Technical Summary

Problems solved by technology

[0003] The multi-channel speech enhancement based on the signal processing method has the advantages of good real-time calculation, but the algorithm design is often based on certain assumptions, such as assuming that the signal is a plane wave propagation, the sound field is uniform and adiabatic, and the noise is stable, etc. There are also some problems in the derivation process of the algorithm. Some mathematical approximations, but this is inconsistent with the actual real scene, and does not take into account the spatial characteristics of the noise reduction space

Therefore, it is difficult to improve the noise reduction performance of these methods in practical application scenarios.

Images