Korean isolated word recognition system using time delay neural network

摘要

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a speech system for recognizing isolated words. In particular, in recognition of isolated words in Korean, learning of words to be recognized and recognition of learned words are performed using a time delay neural network (TIME DELAY NEURAL NETWORK) The present invention relates to a Korean isolated word recognition system using a time delayed neural network so that speech recognition can be accurately performed regardless of the length of utterance of a word.;The conventional speech recognition system using the multilayer neural network can not cope with the variation in the sounding length depending on the speaker even if the same word is used and it is accompanied by the restriction that the recognition result can be obtained only for the sound having the fixed length, There is a problem that it is difficult to recognize a voice having a different vocal length and it is difficult to ensure accurate voice recognition.;The present invention divides a voice signal to be recognized into two demi-syllables (DEMI-SYLLABLE) and integrates a time-delay neural network in which the demi-syllable of the voice signal is input to perform learning and recognition of voice, Time stretching characteristic of the speech recognition system, and thus, it is possible to secure accurate speech recognition regardless of the lingering length corresponding to the inter-speaker speech rate (the pronunciation length of the speech), and is applied to the speech recognition system using the mountain network.