Speech stream detection in non-Gaussian background noise based on statistic characteristics of wavelet coefficient
Liran Shen · Harbin Gongcheng Daxue Xuebao/Journal of Harbin Engineering University · 2004
In practice, non_Gaussian noise is universal and causes serious disturbances, making signal processing difficult. Most strong noises are non_Gaussian, and speech signals are often disturbed and even submerged by strong noises. The traditional algorithm often cannot acquire an ideal effect, which is based on short time energy, short time zero crossing, short time correlation, and short time absolute magnitude difference function. A valid algorithm to detect the speech signal in non_Gaussian background noise was presented according to the statistic characteristics of wavelet coefficient of the speech signal. Speech signal with noise was decomposed by wavelet to investigate the statistic characteristics of wavelet coefficient and different characters were obtained to detect speech signal. Experiments show that the algorithm results in increased correct detection rates and fewer wrong detection rates. The algorithm can steadily resist the influence of noise and can deal with speech signal in real_time. This novel algorithm will be practical when dealing with non_Gaussian noise.