摘要 |
PURPOSE:To shorten the processing time and to increase the work speed by reducing the number of registered words to be actually subjected to pattern collation with respect to an input voice signal as the recognition object. CONSTITUTION:Plural registered words are divided into groups each of which consists of words having the same character string as head parts and are preliminarily set in a main collation dictionary 5, and head parts of several frames of audio signals of registered words are averaged with each group as the unit, and averaged head parts are preliminarily set in an auxiliary collation dictionary 19, and pattern collation between head parts of several frames of an input audio signal and averaged head parts of words registered in the auxiliary collation dictionary 19 is performed in an auxiliary collation means 20, and pattern collation between registered words in the group selected by this pattern collation and the input audio signal is performed in a main collation means 22. |