The speech signal is the most convenient exchange medium, at present, with the aid of the communications system, the people already can make a telephone conversation on the long-distance range freely. However, in environment noise quite big place, if the environment noise has not been processed, has the possibility to cover the speech signal to cause the telephone conversation completely to be unable. At present, the specific noise signal has an effective noise reduction method, but the crowd noise signal, that coincident with the speech signal on the frequency domain, the time domain and feature, there is no effective method of noise reduction. In order to suppress this noise, has conducted the research to the speech signal characteristic, verify speech signal waveform is actually a kind of frequency modulation, amplitude modulation, phase modulation waveform, the carrier is mainly several frequency-stable sine wave, modulation wave is the main carrier of speech information, it is a regular pulse signal. Modulation wave affect the carrier wave, the result is to determine the change way of speech signal crest parameters, to analyze the crest parameters can be restored to the modulation wave. based on wavelet analysis theory, proposes a new analysis method of speech signal-Crest Analysis. In the course of the study, design two programs to practice crest analysis, one for the voice files and another for real-time noise reduction, the experimental results show that the crest analysis can reduce not only simple noise, but also to crowd noises, that coincident with the speech signal on the frequency domain, the time domain and feature. |