ISCA Archive - ASR beyond HMM: a critical review (original) (raw)

ASR beyond HMM: a critical review

Mark A. Huckvale

This paper compares whole-word Hidden Markov Modelling (HMM) with five other techniques of Automatic Speech Recognition (ASR): (i) probabilistic decoding, (ii) sub-word HMM, (iii) dynamic segment modelling, (iv) acoustic-phonetic recognition and (v) adaptive network classification. An example system of each type is referenced and they are described in terms of the source model implied by their construction. Each system is shown to have a different balance between specified and learned constraints.