Browse Prior Art Database

Reading Tutor Using an Automatic Speech Recognition

IP.com Disclosure Number: IPCOM000105535D
Original Publication Date: 1993-Aug-01
Included in the Prior Art Database: 2005-Mar-20
Document File: 4 page(s) / 141K

Publishing Venue

IBM

Related People

Kanevsky, D: AUTHOR [+4]

Abstract

An automatic speech recognition of a speech of a person learning to read is suggested. The peculiarities of an automatic speech recognition in such context are that an actual word that should be spoken is given to a computer in advance. The new task of the speech recognition becomes to recognize whether a student read the given word correctly and recognize his spelling errors. The novelty of this new speech recognition process is the consequence of an unusual speaking manner of the student: pauses between syllables, spelling mistakes, prolonged duration time etc. This requires appropriate modification of basic speech recognition procedures and techniques: labeling, training, decoding, vocabulary, language models, hidden Markov models, neural networks, decision trees (see reff. 1,2,3,4).

This text was extracted from an ASCII text file.
This is the abbreviated version, containing approximately 38% of the total text.

Reading Tutor Using an Automatic Speech Recognition

      An automatic speech recognition of a speech of a person
learning to read is suggested.  The peculiarities of an automatic
speech recognition in such context are that an actual word that
should be spoken is given to a computer in advance.  The new task of
the speech recognition becomes to recognize whether a student read
the given word correctly and recognize his spelling errors.  The
novelty of this new speech recognition process is the consequence of
an unusual speaking manner of the student: pauses between syllables,
spelling mistakes, prolonged duration time etc. This requires
appropriate modification of basic speech recognition procedures and
techniques:  labeling, training, decoding, vocabulary, language
models, hidden Markov models, neural networks, decision trees (see
reff.  1,2,3,4).  Special attention in this disclosure is paid to a
hidden Markov model that incorporates 'break' models (for pauses) in
standard hidden Markov models for speech processing.

      This specific automatic speech recognizer could be used for
developing an automatic reading tutor (for kids, foreigners etc.).
Some features of an interface for such tutor  are described.

      A task of creating an automatic reading tutor requires that the
machine could recognize whether a student read a suggested word
correctly and identify his/her spelling errors.  In principle one can
use an automatic speech recognizer for judging whether a student is
reading correctly orally.  A criteria of the student's ability to
read orally word could be whether the automatic speech recognizer
decode the student's speech correctly.  But the problem is that when
a student (especially a small child) is trying to read orally he/she
distorts significantly a typical speech utterance corresponding this
word: introduces pauses, additional sounds (sighs) etc.  Therefore
standard speech recognizers that are trained on the basis of normal
speech of subjects are not appropriate for the suggested task.  This
poses the problem of designing an automatic speech recognizer that
will be capable of recognition of distorted speech that arises during
oral reading by students.

      An assisting reading device (ARD) can be developed on the basis
of a standard automatic speech recognition modifying the following
blocks: labeling, training, vocabulary, decoding, playback means,
alignment, speech synthesis, user interface.

1.  Labeling.  In addition to labels for standard sounds that are
    produced in a normal speech labeling of sounds characterizing
    oral reading should be done.  These additional sounds that
    require labeling could be sighs, ends of syllables, unusual oral
    production of standard sounds etc.

2.  Training of hidden Markov models requires incorporating
    break-models between sounds.  These breaks appear naturally when
    a student make pauses during oral reading.  For on...