System Request Utterance Detection Based on Acoustic and Linguistic Features

摘要

To facilitate natural interaction for a system such as mobile robot, a new system request utterance detection based on acoustic and linguistic features was employed in this chapter. To discriminate commands from human-human conversations by acoustic features, it is efficient to consider the head and tail of an utterance. The different characteristics of system requests and spontaneous utterances appear on these parts of an utterance. Separating the head and the tail of an utterance, the accuracy of discrimination was improved. Considering the alternation of speakers using two channel microphones also improved the performance. Also we described the system request detection method integrated with a speech recognition system. Boosting was employed as a discriminant method. Its output score is not a probability, though, so the Boosting output score was converted into pseudoprobability using a sigmoid function. The experimental results showed that integration of system request detection and speech recognition improved the performance of request detection. Especially, in the case where 1-best results miss important keywords, the proposed method can recover the keywords from the hypotheses and improve the performance. In the future, we plan to perform experiments using larger corpus and more difficult tasks. In addition, we will investigate a context-dependent approach for request detection. The consideration of new kinds of features is also the assignments.