Research On Speech Signal Processing And Optimizing Based On Compressed Sensing

The traditional Nyquist sampling theory requires that the sampling frequency should not be less than twice the maximum bandwidth of the signal.A small amount of information collected is preserved,while most of the information is ignored,which requires high hardware equipment and creates a waste of resources.With the rapid development of information technology,the Nyquist sampling theory of information collection technology is more and more unable to meet people's requirements for signal processing efficiency.The theory of compressed sensing has solved this problem,as it can be sampled and compressed at the same time.The theory means on the premise of sparsity,observation sequence of signal is got by the signal and measurement matrix product,few of which can restore the high-latitude original signal accurately.Speech signal possesses good compressibility,so Compressed Sensing theory will make the signal compression and reconstruction come true.Different from graphics,research both at home and abroad on the speech signal processing area by Compressed Sensing theory is rarely seen and still at early stage.The paper studies the theory of compression perception and applies the compression perceptual theory to speech signal processing.It describes the implementation process of speech signal compression and reconstruction,introduces the method of reconstructing speech evaluation,and puts forward some optimization and improvement.The concrete work is as follows:Firstly,the paper introduces Compressed Sensing theory,including signal sparse representation,measurement matrix construction and reconstruction algorithm.The paper uses the male and female students in the professional voice library reading aloud voice as the experimental objects.The presentation of BP algorithm and OMP algorithm to speech signal compression and reconstruction is compared and the effects of speech reconstruction because compression ratio and frame length are discussed.The experimental results show that,under the same reconstruction algorithm,the reconstruction quality of male voice is better than that of female voice;for the same experimental speech,BP algorithm is better.Secondly,in the paper,the performance of several common measurement matrix in Compressed Sensing theory to the compression and reconstruction of speech signal is analyzed and compared and some suggestions are given on how to adopt the measurement matrix under different experimental conditions.Experiment shows that compression ratio and frame length are the key factors for the adoption of measurement matrix.On the basis of different compression ratios and frame lengths,measurement matrix will be changed in order to achieve the optimal goal of speech reconstruction.Thirdly,the paper set out from the sparse representation of the signal,and bring in frame algorithm in the redundant dictionary,which can make the signal more sparse and can be compared with the Gaussian random matrix commonly used in the compressed sensing theory to reconstruct the speech quality.According to the experiment,compared with the traditional commonly-used Gaussian random matrix,tight frame performs better during the process of speech reconstruction.Fourthly,Absolute Threshold of Hearing in the psychoacoustics model is applied and it will be helpful to filter out the inaudible and useless signal,reduce the signal non-zero count and increase the signal sparsity in order to improve the quality of reconstructed speech.The experiment demonstrates that the reconstruction of speech will get better.