公开(公告)号：JP2001134284A

公开(公告)日：2001-05-18

申请号：JP31225399

申请日：1999-11-02

Applicant: ATR ADVANCED TELECOMM RES INST

Inventor： YAMAGUCHI TAKESHI ,  KOKIDO SHINGO ,  MOGI YUJI ,  MITSUYA EIJI

IPC: G10L15/06 ,  G10L13/00

Abstract: PROBLEM TO BE SOLVED: To eliminate the trouble at the time of learning a speech recognition device compared with a conventional example and shorten the learning time. SOLUTION: This speech synthesis device 3 selectively speech-synthesizes a plurality of utterance contents of words, sentences, or clauses on the basis of the speech waveform signals of a plurality of speakers, and a parameter learning control part 20 uses the speech-synthesized speech signals of the plurality of speakers and learns, selectively for each kind of the above utterance contents, the parameter of recognition parts of the speech recognition device to be arranged for each of the above kinds. Then, a speech recognition device 4 classifies the inputted speech signals into one of a plurality of prescribed kinds and recognizes them, and on the basis of the speech- synthesized speech signals of the plurality of speakers, the parameter learning control part 20 learns conversion parameters for processing a plurality of feature conversions in feature conversion parts 10-1 to 10-K, and the parameters of a recognition part including a plurality of discriminant functions in discriminant function units 11-1 to 11-K so that the error probability of the speech recognition is minimized.

公开(公告)号：JP2002196780A

公开(公告)日：2002-07-12

申请号：JP2000395001

申请日：2000-12-26

Applicant: ATR ADVANCED TELECOMM RES INST

Inventor： MITSUYA EIJI ,  DABIN PATRICK ,  TAGAWA HIROAKI ,  KOKIDO SHINGO ,  YAMAGUCHI TAKESHI ,  ITO TAKASHI

IPC: G06F3/16 ,  G10L13/06 ,  G10L13/08

Abstract: PROBLEM TO BE SOLVED: To provide a novel communication system utilizing a speech synthesis method. SOLUTION: A signal transmission side apparatus has means for predicting meter parameters of phoneme series corresponding to inputted text and retrieving the combinations of the phoneme units to minimize the distortion when the phoneme units nearest the predicted meter patterns and accumulated in a speech waveform data base are connected and means for forming the waveform connection data consisting of the index information corresponding to the combinations of the retrieved phoneme units and sending the data to a signal reception side apparatus.