Research On Dual-microphone Speech Enhancement Algorithm Based On GS

Microphone array technology is the front-end enhancement technology for speech interaction,it can use the spatial information of signals to suppress noise and improve speech quality.Dual-microphone speech enhancement based on GSC has advantages in noise reduction performance,system volume,and computational complexity,and meets the needs of portable voice interaction devices,therefore it has received extensive attention and application.This dissertation focuses on GSC-based dual microphone speech enhancement,the main contents are as follows:1.Introduces several speech enhancement algorithms based on dual-microphone including GSC,and perform experimental analysis and comparison.It is concluded that GSC has strong ability to deal with directional interference noise and good noise reduction performance with a small number of microphone sensors.On the other hand,it is concluded that GSC has the inherent defects of leakage of the target signal,the poor incoherent noise suppression and poor adaptability to complex acoustic environment.2.Aiming at the defects of insufficient noise suppression capability and leakage of the target signal in complex sound source environment of dual-microphone GSC,an improved dual-microphone GSC speech enhancement based on the target and non-target signal ratio(TNR)is proposed.The algorithm uses the TNR calculated by phase differences to improve the FBF module and ANC module of GSC.Experimental results show that the algorithm can effectively improve the noise suppression ability of GSC based on traditional phase error filter improvement in different environments,and obtain better speech quality in stationary and non-stationary noise environments.However,the algorithm relies heavily on the phase difference information,so the performance tends to become worse in adverse noisy environments.3.Aiming at the defects of the algorithm in the previous chapter,a determinant-based dual-microphone GSC speech enhancement(GSC-DET)is proposed.GSC-DET is based on the determinant analysis of the input correlation matrix,uses voice activity detection and SNR estimation to enhance the output of FBF and improve the accuracy of noise estimation of ANC.This chapter adds a simulation scenario of multiple noise sources.The experimental results show that under the condition of multiple noise sources,the algorithm can effectively reduce the noise components in the output and reduce the distortion of speech,and can flexibly deal with the environment of multiple noise sources.However,the speech processed by the algorithm still has residual background noise.4.Aiming at the defect of the poor incoherent noise suppression and poor adaptability to complex acoustic environment of GSC-DET.an improved dual-microphone GSC speech enhancement method based on ARM and GAN is proposed.The method proposes a T-F-based GAN model as a post-filter module in the GSC structure,and uses Inter-Channel Correlation(ICC)coefficient improve the traditional time-frequency masking.The experimental results show that the algorithm has obvious advantages in adaptability under different types of noise environments.Compared with GSC-DET algorithm,the speech processed by this algorithm is closer to the target speech signal,and the speech distortion is lower.