Browse Prior Art Database

Approximate Word-Spotting Method for Constrained Grammars

IP.com Disclosure Number: IPCOM000113905D
Original Publication Date: 1994-Oct-01
Included in the Prior Art Database: 2005-Mar-27
Document File: 2 page(s) / 36K

Publishing Venue

IBM

Related People

Daggett, G: AUTHOR [+3]

Abstract

Disclosed is a method to allow approximate word spotting mode in a speech recognition system, providing more flexibility for the user.

This text was extracted from an ASCII text file.
This is the abbreviated version, containing approximately 94% of the total text.

Approximate Word-Spotting Method for Constrained Grammars

      Disclosed is a method to allow approximate word spotting mode
in a speech recognition system, providing more flexibility for the
user.

      Speech recognition systems often use grammars to define and
constrain the possible search space.  This reduces the computational
complexity of the task, and generally increases recognition accuracy.

      However, the use of constrained grammars also imposes
restrictions on the user of the system, because the allowable word
sequences are strongly determined.  Small variations in what is being
said, if unanticipated by the grammar, can result in failed
recognition.

      This disclosure presents a method to provide somewhat more
flexibility for the user, while maintaining the advantages of a
restricted grammar for the recognizer.  Assume that grammars can be
written as sentence templates, where states in the template can
include subvocabularies of words.  By allowing selective deletions
and/or arbitrary substitutions in individual grammar components, an
approximate keyword spotting mode can be implemented.

For example, if the full sentence template for a given production is:

1 = show me my < calls | meetings | mail >
then, by allowing selective deletions and/or substitutions:

1 = < show | * > < me | * > < my | * > < calls | meetings | mail >
where "*" indicates an optional deletion or substitution, the user
can legally say:
 show me my calls
 my cal...