Speech recognition method using confidence measure evaluation

• US 6,421,640 B1
• Filed: 09/13/1999
• Issued: 07/16/2002
• Est. Priority Date: 09/16/1998
• Status: Expired due to Term

• Current Assignee: Nuance Communications, Inc. (Microsoft Corporation)
• Original Assignee: Koninklijke Philips Electronics N.V. (Koninklijke Philips N.V.)
• Inventors: Dolfing, Jannes G. A., Wendemuth, Andreas
• Primary Examiner(s): Chawan, Vijay B
• Application Number:
  US09/395,048
• Time in Patent Office: 1,037 Days
• Field of Search: 704/256, 704/232, 704/231, 704/240, 704/255, 704/235, 704/239, 704/243, 704/251, 704/253, 704/242, 704/252, 704/254, 704/257, 704/238, 704/246, 704/236
• US Class Current: 704/236
• CPC Class Codes: G10L 15/10 using distance or distortion measures between unknown speech and reference templates

Abstract

The invention relates to a method of automatically recognizing speech utterances, in which a recognition result is evaluated by means of a first confidence measure and a plurality of second confidence measures determined for a recognition result is automatically combined for determining the first confidence measure. To reduce the resultant error rate in the assessment of the correctness of a recognition result, the method is characterized in that the determination of the parameters weighting the combination of the second confidence measures is based on a minimization of a cross-entropy-error measure. A further improvement is achieved by means of a post-processing operation based on the maximization of the Gardner-Derrida error function.

Claims

1. A method of automatically recognizing speech utterances, the method comprising the steps of:

evaluating a recognition result using an initial confidence measure and a plurality of second confidence measures determined for a recognition result,

determining a resulting confidence measure by combining the initial confidence measure and the plurality of second confidence measures, wherein the determining step includes,

determining parameters for the combining of the second confidence measures using a minimization of a cross-entropy-error measure.

2. The method as claimed in claim 1, wherein the resulting confidence measure is a linear combination of the second confidence measures.

3. The method as claimed in claim 1, wherein in the determination of the parameters weighting the combination of the second confidence measures on the basis of the minimization of the cross-entropy-error measure, a post-processing operation is performed on the basis of maximization of the Gardner-Derrida error function.

4. A method as claimed in claim 1, wherein the combination of the second confidence measures is preceded by a pre-processing operation in which the second confidence measures are formed by multiplication of a plurality third rough confidence measures.

5. The method as claimed in claim 1, wherein the step of determining the modified second confidence measures, every second confidence measure is multiplied by every other second confidence measure, and in that the modified second confidence measures are subjected to said combination for forming the resulting confidence measure.

6. The method of automatically recognizing speech utterances, in which a recognition result is evaluated by means of a confidence measure, as claimed in claim 1, wherein the confidence measure is adapted by means of a user and/or speech utterance-specific offset before comparison with a threshold value serving as the decision limit.

7. The speech recognition system comprising processing units for evaluating a recognition result by means of a method as claimed in claim 1.