Speech Signal Enhancement And Recognition Algorithm

Speech is one of the most important and most important ways for human to transmit information.The main purpose of speech enhancement is to extract the original speech signals from noisy speech signals.The first three chapters mainly focus on speech enhancement,combining wavelet analysis and Variational Mode Decomposition(VMD)algorithm.In this paper,a speech enhancement method based on the combination of VMD and wavelet analysis is proposed to pave the way for the feature extraction and recognition of the next speech signal.In the fouith chapter and the fifth chapter,the MFCC and the logic regression algorithm are introduced in detail.On the basis of the preceding speech enhancement,the MFCC is used to extract the speech feature of speaker at all ages,and then the speech data characteristics of the speaker are classified by the logistic regression method based on the particle swarm.The main research contents include:(1)The problem that the soft threshold and hard threshold are used to denoise the speech signal in the wavelet transform can cause some loss of the original signal.An improved wavelet transform method is proposed,which uses the improved threshold function and adaptive threshold to decompose the wavelet coefficients,and then reconstructs the high-frequency part,so as to reduce the influence of the denoising process on the objective function.(2)A speech enhancement algorithm based on the combination of VMD and wavelet analysis algorithm is proposed.First,the VMD algorithm is used to decompose the original speech signal,and then the correlation analysis is done for each modal component,and the threshold is set.Then,the wavelet threshold denoising of modal components below the threshold is performed,and the useful signals are extracted and reconstructed with after modal components.The simulation results show that the signal to noise ratio after denoising is improved by using the method proposed in this paper,and the partial loss of useful signals in the process of EEMD and VMD denoising is avoided.Finally,a good denoising effect is achieved.(3)The voice features of speakers collected from different ages were analyzed.Finally,the voice features of speakers collected from different ages were extraced through MFCC.(4)Considering the presence of local minima in the traditional logical regression method.In this paper,a classification method of logistic regression based on particle swarm is proposed,which can ease the local minimum problem and improve the convergence speed.This method is used to model and train the voice data collected from all age groups,so as to realize the recognition and classification of speakers of all ages,and analyze the experimental results.