The acoustic processing in the SRI speech-understanding system is described. This system relies heavily upon semantic and syntactic information for understanding the spoken utterance. This information predicts the presence of various words at particular places in the utterance and thereby guides the acoustic processor. The acoustic processor is implemented as a set of word-verification functions which establish the likelihood that a predicted word actually is present at a specified point in the utterance. This approach allows a straight-forward solution of problems of segmentation and within-word coarticulation, which have caused great difficulty in traditional attempts at continuous speech recognition. Typical word verification functions are presented, with illustrations of different methods of analyzing coarticulation effects. The possibility of incorporating these techniques in traditional systems is discussed. Finally, methods of extending these techniques to deal with coarticulation effects which occur across word boundaries are discussed.