| 摘要 |
PURPOSE:To improve the recognition rate of a voice in a very noisy atmosphere by removing the feature quantity of the noise of a speaker playback signal which has large time nonconstancy in the voice band from the feature quantity of a voice containing a noise, and detecting a voice section by a voice section part. CONSTITUTION:A 1st noise removal part 30 subtracts the time spectrum pattern N1 of the noise of the speaker playback signal extracted by a 2nd feature quantity extraction part 20 from the time spectrum pattern X of the voice which is extracted by a 1st feature quantity extraction part 10 and contains the noise to generate a time spectrum pattern after the noise of the speaker playback signal is removed from the voice signal containing the noise. A voice section detection part 40 detects the voice section from the time spectrum pat tern X1 generated by the 1st noise removal part 30 after the noise of the speaker playback signal is removed from the voice signal containing the noise. Consequently, even if the playback sound of a radio set or stereophonic set which has large time nonconstancy is superposed upon the voice signal, the noise is removed properly to realize excellent voice recognition. |
