×

Speech recognition with mixtures of bayesian networks

  • US 6,336,108 B1
  • Filed: 12/23/1998
  • Issued: 01/01/2002
  • Est. Priority Date: 12/04/1997
  • Status: Expired due to Term
First Claim
Patent Images

1. A speech recognition network for inferring parts of speech from acoustic observations having n elements, with a common hidden variable having plural discrete states, comprising:

  • a plurality of mixtures of Bayesian networks (MBNs), each of said MBNs encoding the probabilities of observing the sets of acoustic observations given the utterance of a respective one of said parts of speech;

    each of said MBNs comprising;

    a plurality of hypothesis-specific Bayesian networks (HSBNs), each of said HSBNs encoding the probabilities of observing the sets of acoustic observations given the utterance of a respective one of said parts of speech and given the hidden common variable being in a respective one of its states;

    a combiner which combines outputs of said HSBNs to produce an MBN output of said MBN;

    wherein each one of said HSBNs comprises;

    plural nodes, each of said nodes corresponding to one of said n elements of the acoustic observations, at least some of said plural nodes having dependencies with others of said plural nodes within the one HSBN, a combiner connected to outputs of said nodes, said nodes receiving at their inputs the state of a respective one of the n elements of a current one of the acoustic observations.

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