Research On Microphone Array Adaptive GSC Speech Enhancement Method

Today,with the popularization of intelligence,the application of voice communication is extremely wide,and voice signal processing technology is constantly being updated and promoted.Speech signal processing has high requirements on the quality of speech.However,the speech collected by the speech pickup device is often mixed with a lot of noise,which seriously affects the clarity and intelligibility of the original speech.It is necessary to find an effective method to eliminate the noise to obtain the required voice.Therefore,the importance of voice enhancement technology is particularly prominent.The single-channel speech enhancement method has the advantages of simple principle,low hardware requirements,and easy implementation,but the signal-to-noise ratio of the signal picked up by the mobile sound source is relatively low.The microphone array speech enhancement method can perform timefrequency-spatial joint processing on the picked-up signal,and has superior performance in target signal tracking and anti-interference.This thesis aims to find a better solution for microphone array speech enhancement methods applied to different noise backgrounds.The main research work are included as follows:(1)Aiming at the problem that the traditional generalized sidelobe canceller(GSC)has poor ability to eliminate incoherent noise and the blocking matrix cannot filter out the target signal well under the background of pure gaussian noise,an improved generalized sidelobe canceller speech enhancement method based on post-Fast ICA signal separation is proposed.In this method,an adaptive blocking matrix instead of the traditional blocking matrix is used,which can effectively reduce the leakage of the target signal caused by the unknown direction of the precise wave.And the fast independent component analysis(Fast ICA)algorithm at the output of the GSC is combined to further separate the pure voice and incoherent noise of the output signal,which is used to improve the quality of output voice signal.(2)Aiming at the problem that the noise cancellation ability of traditional generalized sidelobe canceller is greatly attenuated under the background of non-gaussian noise represented by ?-stable distributed noise,an improved generalized sidelobe canceller speech enhancement method based on convex combination filter is proposed.In this method,a convex combination filter is used to replace the traditional adaptive filter,and the unique "parallel" filter of the convex combination is used to realize the adaptive filter's collaborative filtering for ?-stable distributed noise.At the same time,the maximum versoria criterion is selected to improve the cost function of the linear adaptive filtering algorithm,so that the improved adaptive filtering algorithm has stronger robustness against non-gaussian noise,further the improved GSC can effectively eliminate impulse noise and restore a pure target signal.(3)Aiming at the problem that the slow convergence speed and poor performance in noise cancellation of the traditional generalized sidelobe canceller under the background of multiple noises,an improved generalized sidelobe canceller speech enhancement method based on improved spectral subtraction is proposed.In this method,a feedback filter is connected to the output end of the traditional GSC,so that the cost function of the GSC lower branch adaptive filter is changed,thereby the convergence speed of the filter is accelerated.At the same time,an improved spectral subtraction module is connected to the traditional GSC,so that the output voice can be further proceeded,and the residual noise in the output voice is eliminated.Finally the voice signal can be restored with less noise as much as possible.