Single Channel Speech Enhancement Using Fractional Wavelet Transform

The occurrence of various noises attenuates the nature of the original speech signal while recording or transmitting. To enhance the quality and intelligibility of the signals, speech enhancement algorithms are still in demand. For this, different algorithms have been proposed since several years. In this work, fractional wavelet transforms (FrWT)-based speech enhancement method is proposed for quality improvement. Wavelet transform has its inherent property of noise minimization, and in fractional domain, noise filters out without overlapping with the desired signal. The enhanced speech signal is estimated by thresholding the fractional wavelet coefficients. For this, SURE method with hard thresholding is chosen. Simulation results show the efficacy of the proposed method compared to fractional fourier transform (FrFT) and fractional discrete cosine transform (FrDCT). Different subjective and objective results using NOIZEUS corpus show that the proposed method is better than other fractional methods.