Speaker recognition by human requires various elements of anterior cognition of a speaker by another. Automatic recognition takes place only step by step : before trying to isolate a speaker among several possible ones, it is essential to be able to recognize the same person in the variability of his or her own production [Liénard, 1989]. In this case, the recognition system is a speaker verification system. The control of a speaker's variability is a necessary condition for the comparison between several speakers.
We think that in the act of speech, the speaker exerts a personal temporal control over the whole neuro-muscular commands necessary to a sequence of speech sounds. A CVCV sequence can be influenced by the word in which it takes place, by the intonation that carries it, but it keeps the trace of the speaker who produces it in the way by which he programs the sequence of gestures in order to reach the target.
A first study led us to identify the same speaker among voices that were recorded in different conditions. With the help of auditory and acoustic analysis we were able to check the speaker's identity by the observation of formant trajectories and fundamental frequency variation. In a second study we used these criteria to compare two groups of speakers : the first is constituted of particularly close speakers -. same sex, family, .,-, the second of non particularly close speakers.
Cite as: Lhote, E., Abou Haidar, L. (1990) Speaker verification by a vocal proxemy cue. Proc. ESCA Workshop on Speaker Characterization in Speech Technology, 149-154
@inproceedings{lhote90_scst,
author={Elisabeth Lhote and Laura {Abou Haidar}},
title={{Speaker verification by a vocal proxemy cue}},
year=1990,
booktitle={Proc. ESCA Workshop on Speaker Characterization in Speech Technology},
pages={149--154}
}