发明名称 |
METHOD AND SYSTEM FOR RECOGNIZING VOICE |
摘要 |
PROBLEM TO BE SOLVED: To correct voice recognition processing to improve the strength of voice recognition through the existence of an enhanced voice signal. SOLUTION: Enhancement of a voice signal corrects a recognition model learned by using a voice signal not enhanced and an inverse effect in which enhancement of a voice signal causing acoustic mismatching between characteristic data mutually extracted from an enhanced voice signal affects voice recognition performance. Linear prediction coding and cepstral parameter analysis conforming to MEL are compositely performed, and correction is performed at the front edge part of an automatic voice recognition system to calculate cepstral characteristics of a transmitted voice signal used for voice recognition processing by selectively correcting a MEL filter bank in a sense of hearing when the display of a frequency region of an enhanced voice signal is processed. |
申请公布号 |
JPH09179585(A) |
申请公布日期 |
1997.07.11 |
申请号 |
JP19960341738 |
申请日期 |
1996.12.20 |
申请人 |
AT & T CORP |
发明人 |
MAZUIN JII REIMU;JIEI GOODON UIRUPON |
分类号 |
G10L15/20;G10L11/00;G10L15/02;G10L15/08;G10L15/10;G10L21/02;(IPC1-7):G10L3/02;G10L9/14;G10L9/16 |
主分类号 |
G10L15/20 |
代理机构 |
|
代理人 |
|
主权项 |
|
地址 |
|