Research On Robust Microphone Array Beamforming Speech Enhancement Algorithm In Reverberantation Environments

Microphone array signal processing has been widely used in hearing aids,hands-free communication, video conferencing system and robot speech processing system, etc., whose main goals are spatial resolution, interference suppression, and speech enhancement using beamforming techniques. But these techniques has many problems: Firstly, the contradiction between wideband speech signal and traditional narrowband beamforming is becoming more and more prominent, which causing stopband distortion. Secondly, there are usually microphone characteristics errors,such as microphone gain, phase and position uncertainty mismatch errors, which will cause the array response vector distortion and beamformer performance degradation.Finally, when microphone array and speech signal position has a certain distance in a meeting room, microphone array receives this signal is not only speech of directsignal, but also have speech source multiple reflections on the wall, which decreased speech intelligibility. Therefore, we mainly research on robust microphone array beamforming speech enhancement algorithm in reverberantation environment. This thesis does theoretical analysis, computer simulation and experimental induce of the proposed algorithm. Main work as follows:1. In order to solve wideband speech signal beamforming problem, an adaptiveweighted wideband frequency invariant beamforming algorithm based on discrete spatial response variance function is proposed. This algorithm is based on linear constrained minimum variance criterion(LCMV). Firstly, balance matrix of the response function of array space response is defined by binomial calculation of discrete space response deviation function. Secondly, balance matrix is written into the objective optimization function of LCMV in a weighted form. Finally; weighting coefficients in wideband frequency-invariant beamforming(FIB) algorithm are defined as the function of field distance and signal frequency, and the weighting function of signal frequency is dynamic, which updates the weight adaptively.2. In order to solve microphone characteristic error in the practical environment,a robust FIB algorithm based on LCMV diagonal loading is proposed. Firstly, FIB is realized by combining the discrete-space response function with the weighted minimum variance criterion. Secondly, white noise gain is used as the constraint condition to write into the criterion of objective function optimization. Then, White noise gain constraint value is analyzed to reduce the robustness of the microphone characteristic error, which caused by low-frequency signal all-pass filter. Finally,Lagrangian multiplier method and CVX are used to solve the optimal weight vector respectively. Simulation results show that algorithm proposed in this paper has good robust performance, and the low frequency directionality of beamformer is better.3. A modified Wiener filter post beamforming algorithm based on the LCMV in reverberation environment is proposed to suppress reverberation . This algorithm is based on the Wiener post filtering delay and sum beamforming, which has LCMV criterion with Wiener post filtering beamforming. Firstly, assumed that the characteristics of reverberation time in each band are different. The beamforming algorithm is applied to high and low frequency subband domain, and reverberation suppression is improved. Secondly, microphone array receives reverberation signal,which can estimate Wiener weight by spatial information. The results of simulation show that reverberation suppression and the score of speech enhancement system are both improved.4. In view of the actual environment, the intelligibility of speech signals decreased, this paper tests the influence of non-ideal factors of microphone mismatch errors in the anechoic room and car garage through a uniform linear microphone array. The performance of these methods are fully demonstrated with simulations and experiments, which can be used to design robust beamforming in reverberantation environment.