Speech Signal De-noising Research Based On Wavelet Transformation

Speech signal is the important means of transmitting information to each other in daily life.Actually, speech signal is inevitably interfered by the surrounding environment of kinds of noiseand the speech signal quality is also affected. Therefore, it is necessary for the noisy speech signalde-nosing, its main effect is to reduce noise component of the noisy speech signal and enhanceintelligibility of speech. Wavelet analysis has the characteristics of multi-resolution analysis, so it ishelpful for analyzing non-stationary signal, and it can effectively achieve the speech signalde-noising. This paper studies the application of wavelet threshold method where the speechde-noising.First, analyzed the choice of several key parameters in the wavelet threshold method, such aswavelet, wavelet decomposition layers, threshold and threshold function.Secondly, in view of the shortage of traditional threshold functions, this paper studies animproved threshold function. Improved threshold function effectively overcome the disadvantage oftraditional threshold function when the absolute value of the wavelet coefficients less than thethreshold is directly set to zero that cause shocks, while effectively avoiding the absolute value ofthe wavelet coefficients greater than the threshold that lead to the problem of larger deviation. Thesimulation results show that the improved threshold function is better than the traditional thresholdfunction in de-noising effect, and is superior to other threshold functions in signal to noise ratio(SNR) and mean square error (MSE).Again, the low signal to noise ratio, if the wavelet threshold method is used directly to reducethe noise, is not very ideal. Based on this, considered first to use the traditional spectral subtractionof the speech signal preprocessing, and then use the improved wavelet threshold de-noising. Thesimulation results show the effectiveness of the method.Finally, the design of the wavelet threshold method analysis system, the system can choosedifferent key parameters. It can directly reflect the effect in different parameters.