| 摘要 |
<p><P>PROBLEM TO BE SOLVED: To provide a voice interval detecting method and a device capable of accurately detecting a voice interval of a subject person out of tone signals containing noise and voices of persons other than the subject person. <P>SOLUTION: A tone signal is extracted for each frame of predetermined length, and the tone signal of the frame is analyzed to determine whether a voice of a subject speaker is contained in the tone signal of the frame or not. The result of determination is obtained as a voice/non-voice determination value, and a series of recognition units contained in the tone signal and speech production time information for each recognition unit are obtained. The voice/non-voice determination value for each frame that can be obtained in a voice/non-voice determination step, and the series of recognition units and the speech production time information for each recognition unit that can be obtained in a speech recognition step are received, and it is determined for each recognition unit whether speech is produced by the subject speaker or not, on the basis of the magnitude of aggregation results of the voice/non-voice determination values for frames corresponding to the speech production time of the recognition unit. <P>COPYRIGHT: (C)2012,JPO&INPIT</p> |
