Log likelihood ratio normalisation and scoring methods have been studied by many researchers and have improved the performance of speaker identification systems. However, these studies have disadvantages: the recognised distorted speech segments are different for each speaker. Also the background model in log likelihood ratio normalisation is changed in each speech segment even for the same speaker. This paper presents two techniques. Firstly, candidate selection based on significance testing, which designs the background speaker model more accurately. And secondly, the scoring method, which recognises the same distorted speech segments for every speaker. We perform a number of experiments with the SPIDRE database.
Cite as: Kim, J.-H., Jang, G.-J., Yun, S.-J., Oh, Y.H. (1998) Candidate selection based on significance testing and its use in normalisation and scoring. Proc. 5th International Conference on Spoken Language Processing (ICSLP 1998), paper 0261, doi: 10.21437/ICSLP.1998-215
@inproceedings{kim98b_icslp,
author={Ji-Hwan Kim and Gil-Jin Jang and Seong-Jin Yun and Yung Hwan Oh},
title={{Candidate selection based on significance testing and its use in normalisation and scoring}},
year=1998,
booktitle={Proc. 5th International Conference on Spoken Language Processing (ICSLP 1998)},
pages={paper 0261},
doi={10.21437/ICSLP.1998-215}
}