公开(公告)号：JP2000122687A

公开(公告)日：2000-04-28

申请号：JP28614399

申请日：1999-10-07

Applicant: IBM

Inventor： JAMES R LEWIS ,  MARIA E SMITH

IPC: G10L15/06 ,  G10L15/18 ,  G10L15/22

Abstract: PROBLEM TO BE SOLVED: To provide a method for updating a language model in a speech recognition engine for speech application, without subsantially calling a user interactive dialog box during session for correcting erroneous recognition of a speech. SOLUTION: This language model updating method in speech application during a correcting session includes a step for comparing a voice of an original oral statement text with a voice of a substituted text acoustically each other, and a step for updating the language model by correction to update thereby the language model without user interaction, when a result compared acoustically hereinbefore is near enough to indicate correction for mis-recognition error other than editting for a new voice. The updating step is allowed to contain a step for adding a new word to a vocabulary of the speech application.

公开(公告)号：SG79284A1

公开(公告)日：2001-03-20

申请号：SG1999005086

申请日：1999-10-12

Applicant: IBM

Inventor： JAMES R LEWIS ,  MARIA E SMITH

IPC: G10L15/06 ,  G10L15/18 ,  G10L15/22 ,  G10L3/00

Abstract: A method for updating a language model in a speech application during a correction session comprises the steps of: automatically acoustically comparing to one another audio of originally dictated text and audio for replacement text; and, automatically updating the language model with a correction if the acoustical comparison is close enough to indicate that the new audio represents correction of a misrecognition error rather than an edit, whereby the language model can be updated without user interaction. The updating step can comprise adding new words to a vocabulary in the speech application.