SPEAKER ADAPTATION APPARATUS AND SPEAKER ADAPTATION METHOD, RECORDING MEDIUM AND SPEECH RECOGNIZING DEVICE

摘要

PROBLEM TO BE SOLVED: To perform effective speaker adaptation even under a noisy environment.;SOLUTION: Noise data just before the speech section of speech to be used for speaker adaptation are stored into a noise buffer 34. A noise prediction Bayes' learning unit 35 predicts a true speech component in voice data used for speaker adaptation based on the noise data stored in the noise buffer 34, and then determines the distribution of the true speech component. Furthermore, the noise prediction Bayer's learning unit 35 performs the speaker adaptation of a sound model based on the distribution of the true speech component.;COPYRIGHT: (C)2002,JPO