Sinusoidal Modeling Using Elliptic Filter for Analysis and Synthesis of Speech Signals

Abstract

The initial form of sinusoidal model was first introduced in the 1980s. It has been applied to a broad range of speech and audio signal processing such as analysis/synthesis, coding, fundamental frequency modification, time and frequency scale modification. This model uses peaks picking method to estimate the sinusoidal components {amplitudes, frequencies, and phases}. One drawback of this model, however, is that the analysis frame length is generally fixed in analyzing the signal. An analysis frame with fixed length cannot guarantee an optimal spectral resolution to each sinusoidal components. In this paper, in order to obtain better spectral resolution and to estimate the sinusoidal component more accurately, we propose a sinusoidal model using elliptic filter. Experimental results show that the proposed model achieves better performance, in terms of waveform, spectrum, and the synthetic speech quality, than conventional and wavelet transform model.