NONVERBAL UTTERANCE DETECTION DEVICE, NONVERBAL UTTERANCE DETECTION METHOD, AND PROGRAM

摘要

The present invention improves the detection accuracy of a nonverbal utterance. An acoustic model storage unit 10A is configured by a deep neural network with a bottleneck architecture and stores an acoustic model that estimates a phoneme state on the basis of a speech feature. A nonverbal-utterance model storage unit 10B stores a nonverbal utterance model that estimates the posterior probability for the likelihood of nonverbal utterance on the basis of the speech feature and the bottleneck feature. A speech feature extraction unit 11 extracts a speech feature from a speech signal. A bottleneck feature estimation unit 12 inputs the speech feature to the acoustic model and acquires the output from a bottleneck layer of the acoustic model as the bottleneck feature. A nonverbal utterance detection unit 13 inputs the speech feature and the bottleneck feature to the nonverbal utterance model and acquires the posterior probability for the likelihood of the nonverbal utterance output by the nonverbal utterance model.