Acoustic model registration device, speaker recognition device, acoustic model registration processing program and acoustic model registration method

摘要

Acoustic model registration unit can be reliably prevented that the acoustic model recognition performance of the speaker is low is registered, the speaker recognition system, to provide an acoustic model registration processing program acoustic model and registration method. By speaker makes an utterance of the N utterances, microphone 1, you enter N utterances few minutes a speech that was spoken, the voice feature extraction unit 4 is indicative of the acoustic features of speech that has been input this The extracted corresponding to each utterance for each audio feature, the speaker model generation unit 5 generates a speaker model based on the audio feature quantities of N utterances that have been extracted, the checking unit 6, N utterances calculating each degree of similarity between the speaker models created with the voice feature value respectively, only when all similarity N utterances that are calculated is equal to or greater than the threshold, the similarity verification unit 9 creates to be registered in the speaker model database as a speaker model used for speaker recognition speaker model that has been.