An adaptive time-shift analysis for DCT based speech enhancement

Discrete cosine transform (DCT) exhibits different properties from discrete Fourier transform (DFT) due to the lack of phase components. This leads to high variance in DCT coefficients between frames even for perfect stationary inputs. In this paper, an adaptive analysis window-shift is proposed to reduce the drawbacks of the fixed window-shift. The amount of shift in the analysis window is now based on the pitch period, thus increasing the interframe similarities. A number of objective measures are performed to evaluate this proposed technique. The experimental results clearly demonstrate the effectiveness of our technique in improving the DCT based speech enhancement algorithms despite its similarities.