11 Discriminative Keyword Spotting

This chapter introduces a discriminative method for detect ing and spotting keywords in spoken utterances. Given a word represented as a sequence of pho nemes and a spoken utterance, the keyword spotter predicts the best time span of the phonem e s quence in the spoken utterance along with a confidence. If the prediction confidence is a bove certain level the keyword is declared to be spoken in the utterance within the predicte time span, otherwise the keyword is declared as not spoken. The problem of keyword spotti ng raining is formulated as a discriminative task where the model parameters are chosen so the utterance in which the keyword is spoken would have higher confidence than any other spoken utterance in which the keyword is not spoken. It is shown theoretically and empi rically that the proposed training method resulted with a high area under the receiver opera ting characteristic (ROC) curve, the most common measure to evaluate keyword spotters. We pre s nt an iterative algorithm to train the keyword spotter efficiently. The proposed appro ach contrasts with standard spotting strategies based on HMMs, for which the training proced ur does not maximize a loss directly related to the spotting performance. Several expe riments performed on TIMIT and WSJ corpora show the advantage of our approach over HMM-base d alternatives.