Speech Quality Enhancement Based on Sinusoidal Model Using Chebyshev Filter

One shortcoming of the sinusoidal model is that the analysis frame length is generally fixed to a length of about two times or more of the average pitch period. It cannot ensure an optimal spectral resolution to each sinusoidal parameters and results in synthesis artifact such as pre-echo distortion. In this paper, in order to overcome this shortcoming and to obtain the sinusoidal parameters more accurately, a sinusoidal model using Chebyshev filter is initially presented. The performance in terms of the spectrum, phase, and speech quality is then analyzed and compared. Experimental results show that the proposed sinusoidal model achieves better performance than that of conventional model.