Within the framework of an analytical system for speaker recognition, we present a study relating to characteristic information concerning individuals, expressed in short term spectra. The reference base is constituted by spectra representing stable phases of the most discriminatory phonemes, for several dozen speakers. The references are acquired automatically from a number of sentences, in which the selected phonemes appear in low distortion contexts. The speakers are classified by providing the system with the phonemic decomposition of a particular utterance. Once the interesting zones containing useful phonemes have automatically been localized, the resemblance score is calculated using the distances between the corresponding spectra and the references. The results validate the method utilized as soon as the number of useful phonemes is sufficient. Keywords: Speaker identification, Analytic method, Short term spectra information.
Cite as: Bonastre, J.-F., Meloni, H., Langlais, P. (1991) Analytical strategy for speaker identification. Proc. 2nd European Conference on Speech Communication and Technology (Eurospeech 1991), 435-438, doi: 10.21437/Eurospeech.1991-110
@inproceedings{bonastre91_eurospeech,
author={Jean-Francois Bonastre and Henri Meloni and Philippe Langlais},
title={{Analytical strategy for speaker identification}},
year=1991,
booktitle={Proc. 2nd European Conference on Speech Communication and Technology (Eurospeech 1991)},
pages={435--438},
doi={10.21437/Eurospeech.1991-110}
}