Research And Implementation Of Speech Enhancement Algorithm Based On Microphone Array

Speech is the fastest way for people to obtain information.With the rapid development of 5G,speech enhancement algorithm is widely used in video and hands-free conference systems.In these speech communication systems,speech signals are often affected by various noises and echoes,which have a negative impact on communication quality.Therefore,people hope to use speech enhancement algorithms to obtain high-quality speech signals,thereby improving the communication experience.However,traditional single channel speech enhancement algorithms have poor performance in complex environments and limited pickup range.For this purpose,this thesis studies speech enhancement algorithms based on microphone arrays,comprehensively considering the time-frequency and spatial information of speech signals,for sound source localization,using beamforming algorithms and echo cancellation algorithms to enhance the quality of speech signals and effectively transmit information.(1)This thesis proposes an improved sound source localization algorithm to address the issue of inaccurate time delay estimation in generalized cross correlation algorithms under low signal-to-noise ratio conditions.This algorithm obtains more accurate time delay estimation results under low signal-to-noise ratio conditions through an improved phase transformation weighting function,and then performs preliminary direction of arrival estimation.Based on the preliminary direction of arrival estimation results,a search interval is selected,and a joint beamforming algorithm is used for more accurate sound source localization.The experimental results show that the improved sound source localization algorithm can more accurately estimate the direction of arrival,with an average error of approximately 2.6425 degrees.(2)In this thesis,the improved generalized sidelobe elimination beamforming algorithm is directly cascaded with the single channel Wiener filtering method,and the filtering sum beamforming algorithm is applied to the main channel of the generalized sidelobe elimination beamforming algorithm to improve the problem of directional selection of the delay sum beamforming algorithm in the low-frequency part,and the normalized least mean square algorithm is used to improve the convergence speed of the adaptive filter.Finally,the Wiener filtering method is used to enhance the output of the improved generalized sidelobe cancellation beamforming algorithm for single channel speech enhancement.This algorithm not only enhances the suppression effect of non coherent noise,but also improves the suppression ability of residual coherent noise.The experimental results show that the improved beamforming algorithm improves the signal-to-noise ratio by about 8d B and can effectively improve the quality of speech signals.(3)This thesis focuses on the impact of double-talk on echo cancellation in practical hands-free communication processes.The envelope energy algorithm and cross correlation algorithm are combined for double-talk detection,and the normalized minimum mean square algorithm is used for echo cancellation.In the case of double-talk,this algorithm can effectively preserve the near end speech signal and eliminate most echo signals.Using sound source localization,beamforming,and echo cancellation algorithms on hardware platforms for speech signal enhancement processing,experiments have shown that a series of algorithms can effectively eliminate echoes and noise,and improve communication performance.