Browse Prior Art Database

Fast, Accurate Hardware Speech Spectra Pattern Classifier Using No Multiplier

IP.com Disclosure Number: IPCOM000085368D
Original Publication Date: 1976-Mar-01
Included in the Prior Art Database: 2005-Mar-02
Document File: 3 page(s) / 70K

Publishing Venue

IBM

Related People

Dixon, NR: AUTHOR [+2]

Abstract

A system for obtaining a speech spectra pattern classifier has been devised which avoids the use of a multiplier in that system without sacrificing accuracy of speech classification.

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 57% of the total text.

Page 1 of 3

Fast, Accurate Hardware Speech Spectra Pattern Classifier Using No Multiplier

A system for obtaining a speech spectra pattern classifier has been devised which avoids the use of a multiplier in that system without sacrificing accuracy of speech classification.

There are many speech spectra classification methods available to the skilled practitioner in the speech spectra recognition art, but none use an algorithm involving a mean-difference-based normalization. The algorithm employed is:

(Image Omitted)

The particular function of f(3Delta(j)3) found to give appropriate results is: f(3Delta(j)3) = a(i) - Delta(j) over

a(1) - a(2) where Delta is constrained to be

less than a(1), and a(1)>a(2) (2).

This positive linear function of the difference in the mean, in practice, compensates for small differences in the mean between candidate and prototype almost completely, while it allows larger differences to be maintained in the classification. This seems to be a very important and appropriate strategy for speech-spectrum classification.

The figure shows a hardware embodiment of the above algorithm and its particular normalization. Storage is necessary for the spectral prototypes, spectral prototype mean values, and a table g(Delta(j)) where: g(Delta(j)) = f(3Delta(j)3) Delta(j) (3).

Storage in this manner obviates the need for a multiplier.

Timing emanates from a simple clock which is initiated upon receipt of a signal indicating a spectrum for classification and its mean are available. NxM basic clock pulses are required for each classification.

The first step in the recognition procedure is to obtain Delta(j) by diff...