Speech Dereverberation Based On Gammatone Auditory Filter Bank And Inverse Filter By Complex Cepstrum

When using mobile phones, telephones, hearing aids and other equipments inrelatively enclosed indoor space, a strong reverberation phenomenon will appear whenthe location of the sound source is far away from the microphone. The existence ofreverberation will make it difficult to distinguish speeches, declining the quality ofspeech. It seriously impacts the auditory effects. Therefore, we need to find an effectiveapproach to suppress and eliminate the reverberation, reconstructing the original speechin order to enhance speech.This paper modifies the dereverberation algorithm based on complex cepstrum blinddeconvolution. According to the human auditory model, the Gammatone auditory filtersare introduced. One new method based on Gammatone auditory filters and complexcepstrum blind deconvolution is proposed.The main works are as follows. In order to solve the problem of different frequencycomponents affected by reverberation, according to different frequency bands, we takeuse of Gammatone auditory filters to divide speech signal into several sub-band signals.And the complex cepstrum method is used in each sub-band to estimate the roomImpulse Response (RIR) and its inverse, reflecting the difference of band effects causedby reverberation. The estimated inverse of the room impulse response is regarded as theinitial value. Then one adaptive inverse filter with blind deconvolution function isdesigned to achieve the abatement of reverberation. Finally, all the sub-band signals aresynthesized using modified spectral subtraction to enhance the speech, by which furthereliminating later reverberation and background noise and reconstructing speech.At the same time, the processing of sub-bands decomposing based on Gammatonefilters can fit the acoustic model of normal ears well. Independent reverberation process in different sub-bands well reflects the difference of frequency band information, betterretaining the information characteristics in different frequency bands. It also reflects thehuman auditory characteristics well and the final reconstruction speech can achieve abetter sound effects.Both the subjective and objective evaluations are used to evaluate and analyze theresults of the experimental simulation. The proposed algorithm in this paper is comparedwith several other commonly used dereverberation algorithms. The experimental resultsshow that the proposed method obtains a better reverberation effect.