5th International Conference on Spoken Language Processing

Sydney, Australia
November 30 - December 4, 1998

Content-Independent Duration Model on Categories of Voice and Unvoice Segments

Oleg P. Skljarov

Research Institute of Ear, Throat, Nose and Speech, Russia

Trying to understand the experimental data on segmentation of a speech signal by a principle "Voice/Unvoice" has led us to the hypothesis about a pair of logistical dependence between durations of these segments. The segmentation was carried out with the help of the computer program working in quasi real time. The hypothesis about logistic recurrent dependence for sequence of segments durations has allowed to make a conclusion about quasi rhythmical organization of this sequence. With the help of offered recurrent dependencies it is possible to explain statistical peculiarities of speech behaviour of stutterers in comparison with normal speech behaviour. These logistic dependencies were confirmed by direct experimental data. The assumption of origins of specified rhythm is made. These origins are hidden at the level of control of speech production and perception. Is shown, that the chaotic nature of offered dynamics of formation of large-scale temporary structure allows to enter concept of the information into consideration by a natural way.

Full Paper

Bibliographic reference.  Skljarov, Oleg P. (1998): "Content-independent duration model on categories of voice and unvoice segments", In ICSLP-1998, paper 1149.