This paper updates our previous work on the automatic phonetic feature analysis of speech. Previously we have described how a bank of feature-detectors can be used as a front-end to traditional speech recognition pattern matching algorithms, with increased performance in speaker-in-dependent isolated word recognition over purely acoustic front-ends. In this paper we extend the feature analysis to continuous speech: describing the labelling methodology and the additional classification performance of neural network classifiers over the Bayes normal classifier.
Bibliographic reference. Huckvale, Mark A. / Howard, I. S. / Barry, William J. (1989): "Automatic phonetic feature labelling of continuous speech", In EUROSPEECH-1989, 2565-2568.