| 摘要 |
PROBLEM TO BE SOLVED: To decide voice/non-voice with high accuracy, even when environment noise is large around a voice-detecting device and a voice detection program which decide a voice/non-voice section from a voice signal with environment noise superimposed thereon. SOLUTION: A voice characteristic quantity calculating means 1-2 calculates voice characteristic quantities, such as power of a voice signal, a zero crossing rate, peak frequency of a power spectrum and a pitch cycle, with respect to an input signal in each frame given time sequentially from an input means 1-1. A higher order component emphasizing means 1-3 calculates a characteristic quantity obtained by emphasizing the high pass component of a voice frequency characteristic quantity. A voice-deciding means 1-4 performs voice/non-voice decision, on the basis of the characteristic quantities. Furthermore, a means for calculating the existence/absence of the regularity of a voice resonant period is included to perform voce/non-voice decision, on the basis of the existence/ absence of the regularity.
|
