Browse Prior Art Database

Using Alternate Spellings to Generate Baseforms

IP.com Disclosure Number: IPCOM000108464D
Original Publication Date: 1992-Jun-01
Included in the Prior Art Database: 2005-Mar-22
Document File: 1 page(s) / 45K

Publishing Venue

IBM

Related People

Epstein, M: AUTHOR [+7]

Abstract

There are many different algorithms that could be employed to build word models for a speech recognition system. The Tangora Automatic Speech Recognizer builds models for a word that a user wishes to add to the recognizer's vocabulary by utilizing spelling-to-sound rules and one or more pronunciations of the word (1). In certain situations though, it is beneficial for the user to enter a "sounds-like" spelling to help guide these spelling-to-sound rules (2).

This text was extracted from an ASCII text file.
This is the abbreviated version, containing approximately 73% of the total text.

Using Alternate Spellings to Generate Baseforms

       There are many different algorithms that could be
employed to build word models for a speech recognition system. The
Tangora Automatic Speech Recognizer builds models for a word that a
user wishes to add to the recognizer's vocabulary by utilizing
spelling-to-sound rules and one or more pronunciations of the word
(1).  In certain situations though, it is beneficial for the user to
enter a "sounds-like" spelling to help guide these spelling-to-sound
rules (2).

      While it is possible to build models for words using just
pronunciations, using knowledge about how the word is spelled helps
to build more accurate models.  But, sometimes words are not
pronounced as they are spelled.  For example, the user may add "AAA",
but say "triple A".  It is better to use the spelling "triple A" in
the generation of the acoustic model for "AAA" than to use either
just the pronunciation or the incorrect spelling.

      This invention proposes that "sounds-like" spellings be given
by the user when it is wished to provide better data to the
spelling-to- sound rules.  If this "sounds-like" spelling is omitted,
then the true spelling is used.  A "sounds-like" spelling might be
given for a variety of reasons:
           to handle the case where the real spelling is
inappropriate, e.g., "AAA" vs. "triple A";
           to provide hints to the baseform building process, e.g.
"EEOC" vs. "e-e-oh-see";
 ...