| 摘要 |
<p><P>PROBLEM TO BE SOLVED: To provide a device, a program, and a method for easily and accurately discriminating a speaker. <P>SOLUTION: A speaker discrimination device 50 obtains two sets of audio data from respective microphones arranged for two speakers. In addition, the speaker discrimination device 50 frames each of the two sets of audio data. Further, the speaker discrimination device 50 distinguishes, on the basis of a first probability model, whether each of the frames falls into a voiced sound region or an unvoiced sound region. Furthermore, the speaker discrimination device 50 determines whether the distinguished result of the frames falling into the voiced sound region is valid or invalid. At this moment, the speaker discrimination device 50 transforms the energy ratio of the two sets of audio data into a model in which a plurality of probability distributions are mixed, and makes the determination above depending on which of the plurality of probability distributions the energy ratio between the frames falls into. Finally, the speaker discrimination device 50 distinguishes speech regions and silence regions in the two sets of audio data from the distinguished result of the frames after the determination of validity or invalidity, on the basis of a second probability model. <P>COPYRIGHT: (C)2013,JPO&INPIT</p> |
