Publication Details
Discriminative Training Techniques for Acoustic Language Identification
Matějka Pavel, Ing., Ph.D. (DCGM)
Černocký Jan, prof. Dr. Ing. (DCGM)
language identification, language recognition, acoustic modeling, disriminative training, maximum mutual information
This paper presents comparison of Maximum Likelihood (ML)
and discriminative Maximum Mutual Information (MMI) training
for acoustic modeling in language identification (LID). Clear advantage of MMI over ML training is shown. The final error ratecompares favorably to other results published on NIST 2003 data.
This paper presents comparison of Maximum Likelihood (ML)
and discriminative Maximum Mutual Information (MMI) training
for acoustic modeling in language identification (LID). Both approaches are compared on state-of-the-art shifted delta-cepstra features, the results are reported on data from NIST 2003 evaluations. Clear advantage of MMI over ML training is shown. Further improvements of acoustic LID are discussed: Heteroscedastic Linear Discriminant Analysis (HLDA) for feature de-correlation and dimensionality reduction and Ergodic Hidden Markov models (EHMM) for better modeling of dynamics in the acoustic space. The final error rate compares favorably to other results published on NIST 2003 data.
@inproceedings{BUT22218,
author="Lukáš {Burget} and Pavel {Matějka} and Jan {Černocký}",
title="Discriminative Training Techniques for Acoustic Language Identification",
booktitle="Proceedings of ICASSP 2006",
year="2006",
pages="209--212",
address="Toulouse",
url="http://www.fit.vutbr.cz/research/groups/speech/publi/2006/burget_mmi_lid_icassp2006.pdf"
}