First International Conference on Spoken Language Processing (ICSLP 90)
An acoustic-phonetic decoding stage of a recognition system for continuous speech is described, which uses demisyllables as processing units. The task is divided into an explicit localization of the syllable nuclei and their classification based on pat tern-matching techniques and loudness evaluation. The second part classifies the consonant clusters using vowel-context-independent Hidden-Markov-Models with discrete emission probabilities.
Bibliographic reference. Weigel, Walter (1990): "Continuous speech recognition with vowel-context-independent hidden-Markov-models for demisyllables", In ICSLP-1990, 701-704.