×

Method for representing word models for use in speech recognition

  • US 4,903,305 A
  • Filed: 03/23/1989
  • Issued: 02/20/1990
  • Est. Priority Date: 05/12/1986
  • Status: Expired due to Fees
First Claim
Patent Images

1. A method of deriving an acoustic word representation for use in speech recognition systems, said method comprising:

  • creating a word model for each of a plurality of words, each word model having a temporal sequence of acoustic models derived from one or more utterances of its associated word;

    clustering the individual acoustic models from each of the plurality of word models, so as to place individual models into clusters of relatively similar models;

    providing a cluster ID for each such cluster; and

    creating a cluster spelling for a given word, said cluster spelling including a collection of cluster IDs indicating the clusters into which the sequence of acoustic models of said given word'"'"'s word model have been placed by said clustering.

View all claims
  • 0 Assignments
Timeline View
Assignment View
    ×
    ×