Second International Conference on Spoken Language Processing (ICSLP'92)
Banff, Alberta, Canada
In this paper it is shown that the use of third order statistics is useful for voicing decision of a speech signal. We focus our attention in noisy speech contaminated with gaussian and non gaussian noises. Voicing decision based in second order statistics has been widely used with very good results in clean speech and speech contaminated with gaussian noise. We will show that Higher Order Statistics (HOS) are more insensitive to a wide kind of noises, even those produced by sources of periodic noise as car engines. The algorithm has been applied in noise corrupted speech, at different levels of signal to noise ratio, and with different kinds of noise. The results show that this new algorithm gives in all the cases a much better voicing decision than a second order based algorithm.
Bibliographic reference. Moreno, Asunción / Fonollosa, José A. R. (1992): "CUMULANT - based voicing decision in noise corrupted speech", In ICSLP-1992, 531-534.