Research And Implementation Of Speech Enhancement Algorithms Based On Adaptive Beamforming

Speech enhancement technology is a core technology in speech signal processing.which is widely used in multimedia monitoring,modern communication and artificial intelligence systems.Speech enhancement technology can be roughly divided into three categories:single channel speech enhancement,multi-channel speech enhancement and deep learning-based speech enhancement.Multi-channel speech enhancement technology utilizes the time domain and frequency domain of speech signals and the spatial position of microphone arrays to suppress noise effectively.The speech enhancement algorithm based on adaptive beamforming is an important algorithm in multi-channel speech enhancement.The adaptive beamforming algorithm uses the minimum mean square error and other optimal criteria to calculate the weights of each array element in order to enhance the target speech signal and suppress interference and noise.In this thesis,three adaptive beamforming algorithms based on Minimum Variance Distortionless Response(MVDR),Generalized Sidelobe Canceller(GSC)and Robust GSC(RGSC)are implemented and improved.Then the algorithm is used to extract and enhance the target voice in the noisy multi-angle cocktail party experiment The specific research contents of this thesis are as follows:(1)Three kinds of speech enhancement algorithms are analyzed and studied in detail.Based on the near-field far-field model of microphone array and the basic architecture of adaptive beamforming,the theoretical model of speech enhancement algorithm and the hardware and software experimental platform are built.A comprehensive index evaluation model based on subjective and objective combination of speech enhancement effect is proposed.(2)For the adaptive beamforming algorithm with Minimum Variance Distortionless Response,an real-time updating adaptive adjustment strategy for the weights of beamforming filters is introduced in this thesis.The weight of traditional algorithm is calculated according to all the frame information of speech signal it cannot make the noise power of each frame minimum.The improved MVDR weights can be adaptively updated and iterated according to the speech information of a certain number of frames.Experiments show that the improved algorithm can effectively suppress interference and noise.(3)In the traditional Generalized Sidelobe Cancellation algorithm and Robust Generalized Sidelobe Cancellation algorithm,the fixed beam forming module has insufficient ability to suppress interference and noise.In order to solve this problem,this thesis proposes a weight updating MVDR algorithm to replace the fixed sidelobe cancellation algorithm,and to estimate the noise quadratically with the blocking matrix.Experiments show that the improved algorithm has stronger ability to suppress interference and noise than GSC and RobustGSC.In summary,this thesis implements adaptive beamforming algorithm based on MVDR,GSC and RobustGSC.An adaptive adjustment of beamforming weights of the three algorithms is proposed,and a quadratic estimation of noise is made by combining the blocking matrix to improve the algorithm.The experimental results show that the speech enhancement effect of the improved algorithm has been effectively improved,and the effect of interference and noise suppression is good.