On the application of fractional Fourier transform for enhancing noisy speech

Fractional spectral subtraction (FSS), which employs fractional Fourier transform (FRFT), is proposed for speech enhancement Firstly, fractional Fourier transform is applied to the frames of noisy speech samples. The estimated fractional noise spectrum is then subtracted from the derived fractional speech-noise spectrum. Finally, the enhanced speech is obtained from inverse fractional Fourier transform. Theoretic analysis indicates that the new method can find an optimal fractional order, which can best separate speech from noisy signal in the fractional Fourier domain. Thus, the performance of the enhanced speech is effectively improved. Computer simulation shows that the SNR improvement amount and Itakura distance decrease amount of the new method are superior to those of spectral subtraction (SS) for male speech corrupted by additive white Gaussian noise. Moreover, the music noise of the enhanced speech is remarkably suppressed.