Knowledge based approach to consonant recognition

This paper presents a knowledge based approach to consonant recognition. In traditional knowledge based systems, the expert is the linguist/phonetician who attempts to describe and quantify the acoustic events, in the form of production rules into phonetic description. This paper proposes to alter the expert's role so that the expert only needs to provide the basic structure of the phonetic classification. The knowledge itself can then be induced from examples in the agreed structure. Thus the acoustic-phonetic rules are moved from the expert's head to the machine memory via the language of examples rather than via the language of explicit articulation. Recognition results on three broad phonetic classes, namely plosives, semi_vowels and nasals, for a combination of feature sets, for speaker dependent and independent recognition, are presented.