In this paper a method previously suggested by one of the Authors [1], for automatically labelling speech depending upon its speaker discriminating content, is further developed. We show that feature class performance is speaker-dependent, indicating person-specific speech classification to be more appropriate than the original proposal of person-independent classification. A hierarchical classifier is introduced as an alternative to the codebook classifier investigated previously. The ability of person-dependent hierarchical classifiers to automatically distinguish between features with good discriminating properties and those with poor discriminating properties is demonstrated. Finally, those parts of speech that the personalised classifiers label as having good properties for speaker recognition are compared with those given high weightings by an alternative neural network approach. This comparison shows the correlation between the two approaches.
Cite as: Eatock, J., Mason, J.S. (1990) Speaker-dependent speech classification in speaker recognition. Proc. ESCA Workshop on Speaker Characterization in Speech Technology, 94-97
@inproceedings{eatock90_scst,
author={J. Eatock and J. S. Mason},
title={{Speaker-dependent speech classification in speaker recognition}},
year=1990,
booktitle={Proc. ESCA Workshop on Speaker Characterization in Speech Technology},
pages={94--97}
}