Voice source modelling using modified LF model with reduced parameters

Abstract

For many real speech applications such as speech synthesis, speech conversion it is essential to reproduce the voice source waveforms as close as possible to the real ones in order to synthesize a natural speech. In the literature, voice source waveform models such as Liljencrants-Fant (LF) synthesize the voice source waveform using mathematical expressions. Such parametrization produces a poor approximation of real voice source waveforms of the human voice. This paper presents a frequency study of the LF-model. Based on this study, we propose to enrich the spectrum and the phase of the LF-model. An analysis/synthesis scheme was set up to demonstrate the effectiveness of the proposed method. Subjective and objective measures show that the synthesized speech using the proposed method performs better when using the simple LF model.