Browse Prior Art Database

Using an Automatic Speech Recognizer to Improve Speech Ability of Hearing Impaired

IP.com Disclosure Number: IPCOM000121128D
Original Publication Date: 1991-Jul-01
Included in the Prior Art Database: 2005-Apr-03
Document File: 3 page(s) / 140K

Publishing Venue

IBM

Related People

Gopalakrishnan, PS: AUTHOR [+3]

Abstract

It is necessary to provide the hearing impaired or foreign people with feedback on their pronunciation so that they could improve it. Usually such feedback is provided by a teacher's comments or by displaying some particular features of speech (spectrograms, loudness, etc.) via a computer. We suggest a method of using an Automatic Speech Recognizer (ASR) to provide feedback to the hearing impaired or foreign people on the quality of their pronunciation of words.

This text was extracted from an ASCII text file.
This is the abbreviated version, containing approximately 51% of the total text.

Using an Automatic Speech Recognizer to Improve Speech Ability of
Hearing Impaired

      It is necessary to provide the hearing impaired or
foreign people with feedback on their pronunciation so that they
could improve it.  Usually such feedback is provided by a teacher's
comments or by displaying some particular features of speech
(spectrograms, loudness, etc.) via a computer.  We suggest a method
of using an Automatic Speech Recognizer (ASR) to provide feedback to
the hearing impaired or foreign people on the quality of their
pronunciation of words.

      The work for improving pronunciation of the hearing impaired or
foreign people can be effective only if it is performed over a long
enough period and on a regular basis. Therefore, methods that involve
participation of a teacher over the whole course of training is very
expensive.  Also, sometimes it is difficult for a teacher to explain
to a student what kind of error he is producing.  Using computers
that analyze speech can, in principle, give opportunity to users to
train themselves as much as they want (assuming that they have an
access to computers and the cost of computers is less than the cost
of a teacher's time). Therefore, computers that replace (completely
or partially) teachers can significantly reduce the cost of training.
But current computer means for improving pronunciation still have
some disadvantages.  They only provide a display of some particular
features of pronunciation (spectrogram, loudness, intonation, etc.)
so the user can try to control them.  But intelligibility of the
speech depends on many factors as a whole.  It is possible to have
very intelligible speech even when some particular features are
missing (e.g., high frequencies over phone lines).  And it is quite
common that even improving some particular features of speech does
not lead to the improvement of this intelligibility.  The knowledge
of how well the hearing impaired or foreign person was understood by
another person can help to improve their pronunciation.  They can
improve their pronunciation either by trying different variation of
pronouncation of words that were not very well understood (say,
speaking more slowly or carefully) or replacing difficult words by
some synonyms.  Current computer means for improving pronunciation do
not provide such feedback. It is usually provided by a teacher or
another person in the process of conversation.  So the main problem
becomes to find technical means that could replace a teacher in
providing general feedback on pronunciation for a hearing impaired
person or a foreigner.

      The basic suggestion of this article for solving the above
problem is to use ASR as a source of feedback.  The scheme for using
ASR for this special application is as follows.  A person speaks  to
an ASR.  The ASR recognizes speech and produces an output.  The user
finds words that were decoded incorrectly and speaks these words
again, trying to vary hi...