Dismiss
InnovationQ will be updated on Sunday, Oct. 22, from 10am ET - noon. You may experience brief service interruptions during that time.
Browse Prior Art Database

Communication Aid for Hearing Impaired Based on Automatic Speech Recognition

IP.com Disclosure Number: IPCOM000037256D
Original Publication Date: 1989-Dec-01
Included in the Prior Art Database: 2005-Jan-29
Document File: 5 page(s) / 29K

Publishing Venue

IBM

Related People

Bahl, LR: AUTHOR [+5]

Abstract

This article describes a general scheme for adapting the current Automatic Speech Recognizer to the enhancement of speech recognition rate in the hearing impaired.

This text was extracted from a PDF file.
At least one non-text object (such as an image or picture) has been suppressed.
This is the abbreviated version, containing approximately 26% of the total text.

Page 1 of 5

Communication Aid for Hearing Impaired Based on Automatic Speech Recognition

This article describes a general scheme for adapting the current Automatic Speech Recognizer to the enhancement of speech recognition rate in the hearing impaired.

Currently existing technical aids for the hearing impaired only make transformations on spoken acoustical signals before sending them to the sense organs. The full analysis of these signals and the reconstruction of the spoken words is then done by the human brain. The disadvantage of this approach is that the ability of a person with impaired hearing to recognize transformed signals correctly in real time is usually insufficient if these signals are received via touch, sight and other "unnatural" sensors. To overcome these difficulties the present project is based on decoding speech signals and sending to the human sensors not only transformed acoustical data, but labels for decoded parts of speech. These labels can be, for example, the highest ranked words in short lists of words that best match the decoded utterance of acoustic signals, or words from some new vocabulary designed to represent classes of utterances of phonemes that match with decoded part of acoustic signals. Several problems arise in this approach. The first is the synchronization problem: a label for a decoded spoken acoustic segment should be sent before the speaker finishes uttering the word: its latency should be shorter than the reaction time of the receiver.

But in the standard decoding process the decoder only sends a label for a spoken acoustic segment after some delay following the end of that segment. The decoded label should be effectively conveyed to user's brain simultaneously with those acoustic signals that were transmitted undecoded to the user via other channels and perceived by the user through lipreading, hearing aids, etc. This makes it possible for the user to analyze effectively signals that have been transmitted via different channels. Another problem is that current decoders cannot recognize speech accurately enough in any case involving a realistically large vocabulary and continuous speech.

The present invention includes some general principles for overcoming these difficulties as well as some technical details about algorithms that could be useful in this approach. In particular, it will be shown that the phonetic fast match (see
[1]) currently in use can be immediately adapted to solve the synchronization problem and also we show how to modify for this purpose other types of acoustic matches that were developed for automatic speech decoding.

The general block-diagram for the device appears in the figure. It consists of the following blocks:

In block 1 a standard transformation on the incoming acoustic signals is performed (amplification, quantization, etc.). In block 2 the acoustic strings are segmented and the segmented pieces of strings are decoded. In block 3 labels are attached to decoded segme...