Experiments in phoneme recognition using a system of hierarchically organised connectionist networks are reported. The hierarchy of the system is based on the acoustic-phonetic structure of speech. The architecture of each level was designed following the principle that the complexity of the decision boundaries achieved by a multi-layer perceptron classifier depends on the complexity of its architecture. Low complexity classifiers are used to provide two levels of binary feature classification providing four classes of speech signal segments. Two of these classes are then used to provide input into more complex classification procedures for vowels and consonants. The performance of the system in classifying the consonants and vowels in a restricted set of monosyllables is described. Keywords: Phonetic Structure; Speech Analysis; Neural Networks; Speech Recognition
Cite as: Ran, S., Millar, J.B. (1991) Phoneme classification using neural networks based on acoustic-phonetic structure. Proc. 2nd European Conference on Speech Communication and Technology (Eurospeech 1991), 129-132, doi: 10.21437/Eurospeech.1991-28
@inproceedings{ran91_eurospeech,
author={Shuping Ran and J. Bruce Millar},
title={{Phoneme classification using neural networks based on acoustic-phonetic structure}},
year=1991,
booktitle={Proc. 2nd European Conference on Speech Communication and Technology (Eurospeech 1991)},
pages={129--132},
doi={10.21437/Eurospeech.1991-28}
}