A new criterion for speech segmentation based on higher order moments is presented. It is shown that simple estimates of third and fourth order cumulants can provide useful cues for speech segmentation and classification. In fact, the analysis of the cumulants behaviour seems well suited for discriminating voiced, unvoiced and transient sounds. Moreover, the estimated quantities give some indications about the information content of segments of speech. This permit the experimentation of a coding scheme for the speech signal, which has yet been experimented in image coding, founded on a perceptually - based conjecture. Keywords: Higher Order Moments; Speech Segmentation; Speech Classification.
Cite as: Jacovitti, G., Pierucci, P., Falaschi, A. (1991) Speech segmentation and classification using higher order moments. Proc. 2nd European Conference on Speech Communication and Technology (Eurospeech 1991), 1335-1338, doi: 10.21437/Eurospeech.1991-305
@inproceedings{jacovitti91_eurospeech,
author={Gianni Jacovitti and Piero Pierucci and Alessandro Falaschi},
title={{Speech segmentation and classification using higher order moments}},
year=1991,
booktitle={Proc. 2nd European Conference on Speech Communication and Technology (Eurospeech 1991)},
pages={1335--1338},
doi={10.21437/Eurospeech.1991-305}
}