×

Integrated speech recognition and semantic classification

  • US 7,856,351 B2
  • Filed: 01/19/2007
  • Issued: 12/21/2010
  • Est. Priority Date: 01/19/2007
  • Status: Active Grant
First Claim
Patent Images

1. A computer-implemented method performed by a computer with a processor, of training model parameters, the method comprising:

  • identifying a target word sequence from among a set of word sequences, such that the target word sequence has a highest joint association score with a target semantic class, wherein the joint association score is indicative of a correspondence of a semantic class and a word sequence for an acoustic signal, wherein the joint association score incorporates one or more parameters that are applied to one or more features of the word sequence for signal-to-class modeling of the acoustic signal, the one or more parameters including parameters applied to one or more features to match the acoustic signal to the word sequence and parameters applied to one or more features of the word sequence to match the word sequence to a semantic class;

    identifying, with the processor, a competitor word sequence from among the set of word sequences other than the target word sequence, such that the competitor word sequence has a highest remaining joint association score with any available semantic class other than the target semantic class; and

    revising, with the processor, one or more of the parameters to raise the joint association score of the target word sequence with the target semantic class relative to the joint association score of the competitor word sequence with the target semantic class.

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