Research Of Dereverberation Technology Based On Microphone Array

In relatively closed indoor spaces such as conference rooms and studios,the signals that microphones received are the stack of direct sound and multiple reflected sounds,which can produce obvious reverberation effects.Serious reverberation will degrade the performance of different kinds of speech signals processing,such as speech recognition,speech localization,and speech synthesis.With the growing maturity of microphone arrays technology,it is valuable to research the dereverberation algorithm based on the microphone array beamforming technology.After the introducing of theoretical foundation of speech dereverberation,this paper mainly studies the dereverberation method based on complex cepstrum filtering and spectral subtraction.By adopting the generalized sidelobe canceller(GSC)adaptive beamforming technology,an improved GSC-based complex cepstrum filtering and spectral subtraction dereverberation method is proposed.The method of time delay estimation in beamforming is analyzed,and the adaptive eigenvalue decomposition-based time delay estimation is adopted,which is suitable for reverberation speech.The modified algorithm can obtain higher time delay estimation accuracy,and improve dereverberation performanceeffectively.This paper further studies the dereverberation method based on the linear prediction residual cepstrum subtraction.The method initially removes the effect of room impulse response on the basis of cepstrum filtering of the linear prediction residuals of the reverberation speech,and then spectrum subtraction is used to further dereverberate the speech.Combining with transmission function generalized sidelobe canceller(TF-GSC)beamforming technology,an improved dereverberation method based on TF-GSC linear prediction residual cepstrum subtraction is proposed to further reduce the leakage of target signals and improve the adaptive performance and robustness of dereverberation.In this paper,the performance of two improved dereverberation algorithms under different reverberation time is evaluated using subjective and objective quality criterias.The test results show that under the acoustic environment with a reverberation time of 0.6s,for the objective indicators of ?PESQ,?SRNR and ?LPCC,the dereverberation effect based on GSC-CF-SS is increased by 0.265,4.5648 d B and 0.4659 respectively.Furthermore,the dereverberation effect based on TF-GSC-LPRC is increased by 0.315,5.5065 d B and 0.5826 respectively.From the time-domain waveforms and spectrograms,we can see that the two improved methods proposed in this paper both have good dereverberation effects.