A functional approach to the description of speech in terms of phonemic units is exposed. It is based on a structural analysis of phonetically transcribed and syllabified texts, which takes into account the word stress and the syllabic structure of the language. As a consequence, the defined units utilize syllabic microprosodic cues and are present in a limited number of contexts. The method is easily applicable to continuous speech, and substantial improvements in recognizers performance can be expected. Finally, the methodology can be advantageously applied to other speech research areas.
Bibliographic reference. Falaschi, Alessandro (1989): "A functional based phonetic units definition for statistical speech recognizers", In EUROSPEECH-1989, 1013-1016.