Enhancement of Sinusoidal Model by Adaptive-Length Analysis Frame

The sinusoidal model has been applied to speech signal coding, analysis/synthesis, modification, and various other fields. However, one drawback of this model is that the analysis frame size is generally fixed in analyzing the speech signal. As a result, optimal spectral resolution cannot be guaranteed to each sinusoidal component. In this paper, in order to overcome this drawback and to estimate the sinusoidal components more accurately, we propose a sinusoidal model using an adaptive-length analysis frame. In the proposed model, the analysis frame length is varied based on the coarse and refined pitch values. Experimental results have shown that the proposed sinusoidal model can achieve better performance than that of basic and filter bank-based model.