| 摘要 |
PURPOSE:To stably and correctly recognize a voice by providing a discriminative feature quantity extracting part for deriving a discriminative feature quantity for showing a degree of a discriminative feature by a procedure corresponding to each discriminative feature from a spectral component of an input voice and extracting a feature quantity related to a human articulation. CONSTITUTION:A sound signal inputted from a microphone 1 is amplified by an amplifier 2, and inputted to a feature extracting part 3. The sound signal inputted to the feature extracting part 3 is brought to frequency analysis, and based on the obtained, spectral component, a discriminative feature quantity for showing a degree of at least one discriminative feature of a discriminative feature for showing a frictional property, a discriminative feature for showing a voiced sound, a discriminative feature for showing flatness of a spectrum caused by a conversion to a nasal sound, a discriminative feature for showing a buzz property, a discriminative feature for showing flatness of a spectrum of a circular frequency band and a discriminative feature for showing dispersibility of a spectrum is derived by a discriminative feature quantity extracting part 4 by a procedure corresponding to its discriminative feature. Accordingly, a feature quantity related to a human articulation being a discriminative feature quantity can be extracted. In such a way, a voice can be recognized stably and correctly. |
