Coping with disfluencies in spontaneous speech recognition: Acoustic detection and linguistic context manipulation

Nowadays read speech recognition already works pretty well, but the recognition of spontaneous speech is much more problematic. There are plenty of reasons for this, and we hypothesize that one of them is the regular occurrence of disfluencies in spontaneous speech. Disfluencies disrupt the normal course of the sentence and when for instance word interruptions are concerned, they also give rise to word-like speech elements which have no representation in the lexicon of the recognizer. In this paper we propose novel methods that aim at coping with the problems induced by three types of disfluencies, namely filled pauses, repeated words and sentence restarts. Our experiments show that especially the proposed methods for filled pause handling offer a moderate but statistically significant improvement over the more traditional techniques previously presented in the literature.