Study On Speed Noise Reduction Based On Integrated Algorithms

With the continuous development of electronic communication and human-computer interaction,the requirement for speech quality is becoming higher and higher.However,noise can decrease Signal to Noise Ratios and quality of speech signals,so the speech de-noising(enhancing)technology is gradually becoming a research focus in order to improve the quality and meet practical needs.By employing self-adaptive decomposition algorithm as theory basis and in combination with wavelet soft-threshold de-noising method and spectral subtraction method,this paper made a study over de-noising methods of speech signals.The specific study contents are as follows:(1)In consideration of non-stationary characteristic of speech signals,a de-noising method combining Ensemble Empirical Mode Decomposition(EEMD)and wavelet soft-threshold was proposed.Firstly,apply EEMD of self-adaptive decomposing signals to decompose noise-containing speech signals;secondly,based on the characteristics of self-correlation function,determine noise-containing conditions of each intrinsic mode function;then remove noise modes according to scale property of EEMD decomposing;after that,reconfigure the noise-containing mode and use wavelet soft-threshold method to de-noise the reconfigured components;finally,reconfigure the treated components with other mode components to obtain de-noised speech signals.The simulation experiment for this kind of method chose two types of control methods for comparison,in which noise-containing speech signals with eight different Signal to Noise Ratios were simulated;the results of experiments indicate that this method is relatively good in de-noising.(2)In regard to speech enhancement,a method combining Complete Ensemble Empirical Mode Decomposition with Adaptive Noise(CEEMDAN)and spectral subtraction method was proposed.Firstly,use CEEMDAN to decompose the noise-containing speech signals to obtain a series of intrinsic mode functions in descending order from high frequency to low frequency;on the basis of traditional CEEMDAN de-noising method,determine the mode components in need of de-noising treatment by using self-correlation nature;perform de-noising treatment to those components;then reconfigure them to obtain enhanced speech signals.The simulation experiment for this kind of method chose three types of control methods for comparison of speech enhancement results and five types of different noises to compare noise-containing speech signals with different Signal to Noise Ratios.The results of experiments indicate that such method can effectively enhance speech signals and inhibit various types of noises.