A nonlinear probabilistic relaxation labeling for speaker identification is presented in this paper. This relaxation scheme, which is an iterative and parallel process, offers a flexible and effective framework for dealing with uncertainty inherently existing in the labeling of the speech feature vectors. Basic concepts and formulations of the relaxation algorithms are outlined. We then discuss how to model the relaxation scheme to the labeling of the speech feature vectors for the speaker identification task. The implementation is tested on a commercial speech corpus TI46. The results using several codebook sizes obtained from the proposed approach are more favorable than those from the conventional VQ (Vector Quantization)-based method.
Cite as: Pham, T., Wagner, M. (1998) Speaker identification using relaxation labeling. Proc. 5th International Conference on Spoken Language Processing (ICSLP 1998), paper 0949, doi: 10.21437/ICSLP.1998-232
@inproceedings{pham98_icslp,
author={Tuan Pham and Michael Wagner},
title={{Speaker identification using relaxation labeling}},
year=1998,
booktitle={Proc. 5th International Conference on Spoken Language Processing (ICSLP 1998)},
pages={paper 0949},
doi={10.21437/ICSLP.1998-232}
}