Method for likelihood computation in multi-stream HMM based speech recognition

• US 7,480,617 B2
• Filed: 09/21/2004
• Issued: 01/20/2009
• Est. Priority Date: 09/21/2004
• Status: Expired due to Fees

• Current Assignee: Nuance Communications, Inc. (Microsoft Corporation)
• Original Assignee: International Business Machines Corporation
• Inventors: Chu, Stephen Mingyu, Goel, Vaibhava, Potamianos, Gerasimos, Marcheret, Etienne
• Primary Examiner(s): MCFADDEN, SUSAN IRIS
• Application Number:
  US10/946,381
• Publication Number:
  US 20060074654A1
• Time in Patent Office: 1,582 Days
• Field of Search: 704/256
• US Class Current: 704/256
• CPC Class Codes: G10L 15/144 Training of HMMs

Abstract

A method for speech recognition includes determining active Gaussians related to a first feature stream and a second feature stream by labeling at least one of the first and second streams, and determining active Gaussians co-occurring in the first stream and the second stream based upon joint probability. A number of Gaussians computed is reduced based upon Gaussians already computed for the first stream and a number of Gaussians co-occurring in the second stream. Speech is decoded based on the Gaussians computed for the first and second streams.

Claims

1. A method for speech recognition, comprising the steps of:

determining active Gaussians related to a first feature stream and a second feature stream by hierarchically labeling at least one of the first and second streams by surveying Gaussians in multiple resolutions;

determining active Gaussians co-occurring in the first stream and the second stream based upon joint probability wherein the first stream includes an audio stream and the second stream includes a video stream;

reducing a number of Gaussians computed for the second stream based upon Gaussians already computed for the first stream and a number of Gaussians co-occurring in the second stream; and

decoding speech based on the Gaussians computed for the first and second streams by employing multi-stream hidden Markov models.