| 摘要 |
PROBLEM TO BE SOLVED: To recognize a telephone voice in a short time in simple configuration by providing a detection step for detecting an audio domain, a determination step for determining an acoustic model set and a recognition step for recognizing the voice in the audio domain while using the determined acoustic model set or the like. SOLUTION: The audio domain is detected from a voice waveform while using power information or the like (S202). Next, a digitized acoustic signal is analyzed in the detected audio domain and an acoustic characteristic parameter such as cepstrum time sequence is outputted (S203). Next, the class of telephone or line is discriminated while using analytic data in a non-audio domain before the audio domain between non-audio domains before and after the audio domain detected in S202 (S204). When acoustic data from a portable telephone are discriminated as a result of this discrimination, S205 is started and when acoustic data from the other telephone are discriminated, S206 is started. Then, the audio domain part is recognized while using an acoustic model for portable telephone (S205). |
