The Research Of Speech Enhancement Based On Fractional Fourier Transform Domian And Time Domain

Due to the development of human-computer interaction, speech signal plays an important role in daily life. But the speech signal is affected by the random noise, sometimes even will cause the failure of the voice system which influence people working life. Therefore, to deal with the noise and improve the communication is particularly important to the quality. Fourier transform is rapid development due to its time domain and requency domain of speech enhancement in a wide range of applications, This paper mainly based on speech enhancement algorithm in FRFT to discuss the advantages and the disadvantages of research and make the improvement. So the main completed work as follows:First, this paper will promotion traditional spectral subtraction in frequency domain to the fractional order domain spectral subtraction. So this papers use the MMSE as the basis, select the best transform order times. Second, considering the actual engineering application, the multichannel sampling and cycle non-uniform sampling is widely applied to, but according to the nature of the fractional order, this paper used formula about multichannel sampling and cycle the spectrum characteristics of the sampling uniformity and rebuild which proposed by Tao R, Li Z B, Wang Y to look for a method of interpolation and reconstruction. Third, considering the limitation of endpoint detection method, this paper is proposed three state machine model based on the fractional domain, output in the best judgment in moving average filter first order differential output as a discriminant, using some historical frame speech information judgment, enhance the end-point detection performance in different SNR environment of stability. Four, for fractional order spectral subtraction of the remainder of the music noise problems didn’t get the solution of fundamentally, combined with the noise environment changes, in the fractional order domain smoothly, in order to ensure that the spectrum continuity, effectively reduce the influence of "music noise".At last, with different SNR of the speech signal on the basis of theoretical analysis in this paper, the paper to get the enhanced speech signal. The results show this paper puts forward the improving method in the fractional order based on the spectral subtraction, not only can adapt to the low SNR environment and the influence of the sudden noise. Still, it can inhibit the generation of further music noise and has a good effect so as to improving the effectiveness and practical applicability.