Noise reduction using modified phase spectra and Wiener Filter

In this paper our aim is to investigate a modified Wiener filter for the enhancement of speech that has been corrupted by fluctuating noise. In real noise environment, where the SNR of speech maybe fluctuating about 5 to 10dB, because typical speech enhancement methods always estimate the noise power spectrum directly by the average noise power spectrum of the pauses in speech, but there still some error and it is impossible to track the noise in speech interval. In order to settle this problem, our proposed method estimates the noise spectrum in the form of Posteriori SNR according to its Gaussian statistical model and track the noise spectrum form degraded speech directly and changes at every frame in the noise reduction processing of speech. Then use an iterative structure contains noise estimation, modified wiener filter and phase spectrum. Finally, the performance of the proposed method was tested in 4 kinds of fluctuating noises, and resulting in improved results over classical MMSE algorithms at the low fluctuating SNR.