Speech recognition system wherein the consonantal characteristics of input utterances are extracted

摘要

In a speech-recognition system in which time-series patterns of characteristic quantities are extracted from an input utterance within a voiced interval, the localpeak patterns, which are the results obtained by stably extracting the characteristics of the steady vowel portions, and the consonantal patterns, which are the results obtained by stably extracting the characteristics of the consonantal portions, are recognized and identified. This makes it possible to perform accurate and stable speech-recognition including also the characteristics of the consonants.