×

Method for deriving acoustic models for use in speech recognition

  • US 4,914,703 A
  • Filed: 12/05/1986
  • Issued: 04/03/1990
  • Est. Priority Date: 12/05/1986
  • Status: Expired due to Fees
First Claim
Patent Images

1. A method of deriving an acoustic model of a first class of speech sounds, which acoustic model is to be compared against a portion of speech to be recognized to determine the likelihood that the portion of speech corresponds to that first class of speech sounds, said method including:

  • calculating a first statistic of the first class of speech sounds from acoustic data derived from one or more samples of that class of speech sounds;

    calculating a second statistic from acoustic data derived from samples of a second class of speech sounds, at least some of which samples correspond to speech sounds which do not belong to the first class of speech sounds for which said acoustic model is being made;

    calculating a combined statistic which is a weighted combination of the first and second statistics; and

    using said combined statistic as at least a part of said acoustic model of said first class of speech sounds.

View all claims
  • 1 Assignment
Timeline View
Assignment View
    ×
    ×