Study On Key Technologies Of Speech Processing In Vehicle Environment

With the development of science,technology and economy,voice control system has become an indispensable part of vehicle equipment.The voice commands received by the vehicle voice control equipment are disturbed by the noise of driving environment to varying degrees,and even cause the failure of the vehicle voice control equipment.Therefore,vehicle voice processing technology has become the research focus in the field of artificial intelligence.The limitations of traditional speech enhancement algorithms in vehicle environment are analyzed,the basic principles and characteristics of traditional speech enhancement algorithms are expounded,the problems of traditional algorithms are experimentally studied,and then a speech enhancement algorithm based on non-negative matrix decomposition is proposed to be applied in vehicle environment,and further research on this algorithm is carried out in this paper.Aiming at the limitation of single channel non-negative matrix speech enhancement algorithm in frequency domain,this paper focuses on the non-negative matrix speech enhancement algorithm based on complex frequency domain and convex hull convolution,and processes the noisy speech received in driving environment.Firstly,the traditional complex nonnegative matrix algorithm is improved by using KL divergence model.A speech enhancement model based on complex nonnegative matrix decomposition of KL divergence model is proposed.The modulation spectrum of traditional phase spectrum compensation algorithm is modified by using frequency domain consistency constraints.Secondly,a speech enhancement model based on convex hull clustering convolution non-negative matrix decomposition based on discrete wavelet packet transform is proposed.The speech signal is processed in time domain using discrete wavelet packet transform,which overcomes the speech distortion caused by short-time Fourier transform,improves the convex hull convolution non-negative matrix decomposition algorithm,and applies it to speech enhancement in vehicle environment.The simulation results verify the effectiveness of the two methods.Finally,two multi-channel speech processing models are proposed to solve the application problems in the proposed vehicle environment.The first method is unsupervised multi-channel speech enhancement method based on non-negative matrix decomposition that is using non-negative matrix decomposition algorithm to construct blocking matrix module of beamforming and effectively avoiding the shortcomings of blocking residual and expected signal cancellation caused by inappropriate blocking matrix construction.The second method is the speech source separation and enhancement method based on statistical model.Non-negative matrix decomposition algorithm is introduced into speech enhancement algorithm based on statistical model and generalized cross-correlation delay estimation modelrespectively.A new angle spectrum function is constructed to separate and enhance noisy speech signals on vehicle without using prior information.The experimental results show that under the actual vehicle noise interference,the proposed algorithms can reduce or suppress the vehicle noise interference,reduce distortion,and improve speech quality and intelligibility.