We present a new algorithm for speaker recognition (the Sequential Non-Parametric system, or SNP) that has the potential to overcome two limitations of the current approaches. It uses sequences of frames instead of one frame at a time; and it avoids the need to model a speaker with mixtures of Gaussians by scoring the data non-parametrically. Although at an early stage in its development, SNP's output can be interpolated with that of our GMM system to outperform state-of-the-art GMM's. Comparative results are presented for the 1998 NIST Speaker Recognition Evaluation test set.
Cite as: Corrada-Emmanuel, A., Newman, M., Peskin, B., Gillick, L., Roth, R. (1998) Progress in speaker recognition at dragon systems. Proc. 5th International Conference on Spoken Language Processing (ICSLP 1998), paper 1017, doi: 10.21437/ICSLP.1998-245
@inproceedings{corradaemmanuel98_icslp,
author={Andres Corrada-Emmanuel and Michael Newman and Barbara Peskin and Lawrence Gillick and Robert Roth},
title={{Progress in speaker recognition at dragon systems}},
year=1998,
booktitle={Proc. 5th International Conference on Spoken Language Processing (ICSLP 1998)},
pages={paper 1017},
doi={10.21437/ICSLP.1998-245}
}