This paper proposes an integration of the time-alignment information in the decision making for HMM-based text-dependent speaker verification. The principle is to consider acoustical score and time-alignment as joint observations for which a log-likelihoodratio is computed and compared to a threshold. It is shown that such integration has two distincts aspects, one being a kind of adaptation of the acoustical score threshold to the observed alignment, the other being the integrationof the speaker-specificity information of the alignment in the decision making. Exper-iments on a large-scale and realistic database are reported. They showthe interest of the proposed method and encour-age further investigation in such an approach.
Cite as: Charlet, D. (1999) Integrating time-alignment information into the decision making for text-dependent HMM-based speaker verification. Proc. 6th European Conference on Speech Communication and Technology (Eurospeech 1999), 1967-1970, doi: 10.21437/Eurospeech.1999-433
@inproceedings{charlet99_eurospeech,
author={D. Charlet},
title={{Integrating time-alignment information into the decision making for text-dependent HMM-based speaker verification}},
year=1999,
booktitle={Proc. 6th European Conference on Speech Communication and Technology (Eurospeech 1999)},
pages={1967--1970},
doi={10.21437/Eurospeech.1999-433}
}