The perceptual effects of aliasing distortion in glottal flow modelling

Abstract

When modelling the glottal flow signal in a discrete- time system, the aliasing distortion that is produced is typically ignored. The assumption is that the percep- tual effects are negligible if the sampling frequency is sufficiently high. In this paper, we implement a recently developed version of the LF model, which eliminates aliasing distortion. By comparing it to the standard application of the LF model, which introduces aliasing distortion, we can explore if, and to what extent, aliasing distortion is perceptible in the modelled voice source signal. The results of a listening test demonstrate that the aliasing distortion is almost always perceptible, even when the sampling frequency is relatively high. The perceptual artefacts are very noticeable when the fundamental frequency is high, particularly in combination with a tense voice quality. In conclusion, therefore, if high-quality modelling of the glottal flow signal is required, aliasing-free source modelling is recommended.