Three cross-language ASR experiments which use hidden Markov modelling are described. Their goal is to process the signal so that it better exploits its linguistically relevant properties for consonant identification. Experiment 1 shows that consonant identification improves when vowel transitions are used. Particularly the consonants' place of articulation is identified better, because the vowel transitions contain formant trajectories which depend on the consonant's place of articulation. Experiment 2 shows that mapping acoustic parameters onto phonetic features before applying hidden Markov modelling greatly improves consonant identification rates. In experiment 3, the acoustic parameters from the vowel transitions are also mapped onto consonantal (not vocalic!) features ("relational processing"), as are the acoustic parameters belonging to the consonants. The additional use of vowel transitions does not further improve consonant identification, however. This is probably due to undertraining of the vowel transitions in the Kohonen network.
Cite as: Koreman, J., Barry, W.J., Andreeva, B. (1998) Exploiting transitions and focussing on linguistic properties for ASR. Proc. 5th International Conference on Spoken Language Processing (ICSLP 1998), paper 0548, doi: 10.21437/ICSLP.1998-651
@inproceedings{koreman98b_icslp,
author={Jacques Koreman and William J. Barry and Bistra Andreeva},
title={{Exploiting transitions and focussing on linguistic properties for ASR}},
year=1998,
booktitle={Proc. 5th International Conference on Spoken Language Processing (ICSLP 1998)},
pages={paper 0548},
doi={10.21437/ICSLP.1998-651}
}