VERFAHREN ZUR ERMITTLUNG VON MERKMALEN UND SPRACHERKENNUNG UND ERKENNUNGSEINRICHTUNG.

摘要

The present invention relates to a method and an apparatus of extracting the feature of voice and a method and apparatus of recognizing the voice by using the feature, in which a time-space pattern of scalar field obtained from the voice signal and regulated by the time axis and the space axis is converted into a vector field pattern by the space differentiation, whose orientation parameter is quantized to obtain a two dimensional pattern as the feature of voice by utilizing the quantization value. Thereby, a high recognition rate can be obtained when recognizing a large vocabulary and in case of a recognition by a non-specified speaker.