The problem of how to train variance parameters on scarce data is addressed in the context of text-dependent, HMM-based, automatic speaker verification. Three variations of variance flooring is explored as a means to prevent over-fitting. With the best performing one, the floor to a variance vector of a client model is proportional to the corresponding variance vector in a non-client multi-speaker model. It is also found that creating a client-model by adapting the means and mixture weights from the non-client model while keeping variances constant works comparably to variance flooring and is much simpler. Comparisons are made on three large telephone quality corpora: Gandalf, SESP and Polycost.
Cite as: Melin, H., Koolwaaij, J.W., Lindberg, J., Bimbot, F. (1998) A comparative evaluation of variance flooring techniques in HMM-based speaker verification. Proc. 5th International Conference on Spoken Language Processing (ICSLP 1998), paper 0467, doi: 10.21437/ICSLP.1998-248
@inproceedings{melin98_icslp,
author={Haakan Melin and Johan W. Koolwaaij and Johan Lindberg and Frédéric Bimbot},
title={{A comparative evaluation of variance flooring techniques in HMM-based speaker verification}},
year=1998,
booktitle={Proc. 5th International Conference on Spoken Language Processing (ICSLP 1998)},
pages={paper 0467},
doi={10.21437/ICSLP.1998-248}
}