It is a basic problem how to construct an effective code book discriminating phoneme classes. Especially, if various acoustic. features are used, they are mutually related and also have context dependent characteristics. In this paper, an optimal code book is obtained by clustering various acoustic features by maximizing the mutual information between the acoustic and phoneme categories. The main issue of this paper is the consideration of context in the hierarchy of acoustic features.
Cite as: Shirai, K., Kitagawa, E., Endo, T. (1991) Optimal construction of context sensitive quantizer for phoneme recognition in continuous speech. Proc. 2nd European Conference on Speech Communication and Technology (Eurospeech 1991), 405-408, doi: 10.21437/Eurospeech.1991-103
@inproceedings{shirai91b_eurospeech,
author={Katsuhiko Shirai and E. Kitagawa and T. Endo},
title={{Optimal construction of context sensitive quantizer for phoneme recognition in continuous speech}},
year=1991,
booktitle={Proc. 2nd European Conference on Speech Communication and Technology (Eurospeech 1991)},
pages={405--408},
doi={10.21437/Eurospeech.1991-103}
}