Application of adaptive neural networks to speech recognition

This paper is concerned with the application of adaptive neural networks (ANNs) to the automatic classification of speech segments. Speech segments as used here will be taken to mean short segments on the order of dyads, half-syllables, or possibly syllables. At this time it is an open question as to the type of subword segments best employed in speech recognition systems. To the writer, a dyad segment appears to be the best candidate. For the present only pre-segmented data will be investigated. It should be mentioned, however, that segmentation and classification of segments are not independent and thus a full fledged recognition system would probably entail feedback from the segment classifier to the segmenter.