Speech Enhancemen Based On The Fractal Theory

what is actually happening, speech signal is always interfered by some noise which is from surrounding architectures, transmission media, electrical facilities and so on. Sometimes, the surrounding noise effects so heavily that the speech signal can not be distinguished. The lower quality of speech signal results in serious deterioration of speech signal processing system. Now pretreatment of speech signal plays an very important role to improve the performance of system. we mainly discussed the application of fractal method de-noising of speech signal.As a multidisciplinary and comprehensive subject, digital speech signal processing has become more and more important nowadays. Due to the non-linear characteristics of speech signals, the performance of traditional techniques based on linear methods cannot be improved any more. The lately developed and ameliorated non-linear theories have brought new directions for speech signal processing. At the same time, people have showed great concern for the applications of fractal method in speech processing.In this paper, we firstly introduced the basic properties of speech organs, pronouncing mechanism and the characteristics of Mandarin speech signal. In addition, we introduced the basic techniques in speech analysis and processing and their shortcomings for comparison.Secondly, after analyzing the advantage of wavelet showed in the time-frequency representation of speech signal, we discussed the application of wavelet in speech de-noising to different kinds of noise. The experimental results showed that wavelet transform can effectively eliminate different noise under different SNR.At last, after analyzing the transfer characteristic of the speech signal and the random noise on wavelet transform in multi-resolution, a novel de-noising algorithm based on fractal dimension and wavelet threshold is presented. The characters and fractal dimensions of noise and signal on wavelet transform are analyzed. Compared with others, the processes of the algorithm has some advantages such as efficiently constructing, effectively wiping off the noise, easily programming. Examples prove that the algorithm has better de-nosing performance.