| 摘要 |
PURPOSE:To exactly recognize to which of comparison reference voices recorded in advance a voice uttered by an unspecific speaker is the nearest. CONSTITUTION:An uttered voice is expressed as a voice curved surface in a coordinate space consisting of an acoustic pressure axis, a time axis and a frequency axis, and the voice curved surface concerned is converted to a feature curved surface by using a conversion coefficient for executing the conversion so that a difference between the maximum value and the minimum value of the curved surface becomes a value determined in advance, at every coordinate axis. The coordinate space is subdivided into minute spaces, the feature curved surface is applied to the subdivided coordinate space, and to each of the minute spaces in which the feature curved surface concerned exists, a weight value is allocated at every word, and an assembly of this weight value is compared with a voice to be recognized at the time of recognizing a voice and sets it as comparison reference voice feature information which becomes a reference for specifying a voice. The feature curved surface of the voice to be recognized is applied to the sub-divided coordinate space, and by adding up the weight values of the minute spaces containing the feature curved surface, the similarity is calculated. |
