Browse Prior Art Database

Template Selection Method for Speaker-Independent Word Recognition

IP.com Disclosure Number: IPCOM000061559D
Original Publication Date: 1986-Aug-01
Included in the Prior Art Database: 2005-Mar-09
Document File: 2 page(s) / 50K

Publishing Venue

IBM

Related People

Watanuki, O: AUTHOR

Abstract

A method is proposed to select optimal templates for speaker-independent word recognition when there is confusion between clusters of different word categories. Conventional template selection methods for speaker-independent word recognition do not consider the effect of a neighboring word category [1, 2, 3]. The method described in [4] optimizes the templates by repeating recognition and removes the templates that are harmful for recognition. However, this method is time consuming. The proposed method makes the optimal selection of templates from the word categories with confusion in the following way: (1) Compute distance matrix by merging the utterances belonging to two or more categories with some confusion. The distance is normally computed by DP matching.

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 53% of the total text.

Page 1 of 2

Template Selection Method for Speaker-Independent Word Recognition

A method is proposed to select optimal templates for speaker-independent word recognition when there is confusion between clusters of different word categories. Conventional template selection methods for speaker-independent word recognition do not consider the effect of a neighboring word category [1, 2,
3]. The method described in [4] optimizes the templates by repeating recognition and removes the templates that are harmful for recognition. However, this method is time consuming. The proposed method makes the optimal selection of templates from the word categories with confusion in the following way: (1) Compute distance matrix by merging the utterances belonging to two or more categories with some confusion. The distance is normally computed by DP matching. (2) Perform clustering using the distance matrix computed above. Any clustering algorithm for template selection may be used for this purpose. (3) If the word categories are correctly separated by the clustering algorithm, the selected templates are considered to be optimal and are used for recognition. (4) If the word categories are not correctly separated, the templates are selected by the following criteria: (i) If the members of a cluster are from a single category, the selected representative of the cluster is used for a template. (ii) If the members of a cluster are from more than two categories, the separation factor,
S. with respect to a vector, v, is computed according to the definition below. The vector in one category that maximizes the separation factor is selected as a template for the first category.

(Image Omitted)

The templates for...