ISCA Archive ECST 1987
ISCA Archive ECST 1987

A new approach to acoustic-phonetic decoding by means of hidden-Markov-modelling of speech

B. Keck

A connected speech recognition method based on stochastic modelling of speech is presented. The speech signal is segmented and labelled by a Viterbi-algorithm that finds the optimal path in an acoustic-phonetic Hidden-Markov-Model (HMM) of the language. This HMM is an integration of speech subunit models and a language model on the subunit level.