Study Of Speech Enhancement Technology Under Low SNR

ABSTRACT:Humans can distinguish and follow the interested speech signal in several speakers’situation, which is a unique human ability for understanding. In the machine speech processing area, how to separate the original speech signal from the environmental noise is a very important topic. Most speech enhancement methods get better result with the stable noise under high SNR environments, such as Wavelet thresholding method, but they show poor performance if the speech signal polluted by non-stationary noise with lower SNR.In this thesis, the enhancement algorithms of noisy speech signal with non-stationary noise under lower SNR are studied. First, a de-noising method combined minimum mean-square short-time log-spectral amplitude (MMSE-LSA) with wavelet threshold is proposed can improve the voice signal to noise ratio effectively and enhanced the intelligibility of speech signal. Second, the enhanced voice from the method combined MMSE-LSA with wavelet threshold saved as a priori information used in Kalman filtering speech enhancement method.In this way the effect of enhanced voice denoising has been further improved. And the performance of the algorithm is verified by simulation.Simulation results show that the proposed algorithm could remove noise in the speech signal effectively and improves the practicability.