The aim of this paper is to find a reliable function using Time-Frequency Representations (TFR) for epoch detection, making no assumptions about the speech production model. This is achieved by adapting the optimum time-frequency detector to the problem of glottal closure instant (GCI) determination. An ideal function for such purpose may show only one maximum per period. However, since this situation is not assured because secondary maxima may appear, we also propose a post-processing method based on morphological filtering.

Peer Reviewed