Correcting substitution errors during automatic speech recognition by accepting a second best when first best is confusable

• US 8,600,760 B2
• Filed: 11/28/2006
• Issued: 12/03/2013
• Est. Priority Date: 11/28/2006
• Status: Expired due to Fees

• Current Assignee: GM Global Technology Operations LLC (General Motors Company)
• Original Assignee: General Motors LLC (General Motors Company), GM Global Technology Operations LLC (General Motors Company)
• Inventors: Grost, Timothy J., Chengalvarayan, Rathinavelu, Clark, Jason W., Abeska, Edward
• Primary Examiner(s): Smits, Talivaldis Ivars
• Assistant Examiner(s): Kazeminezhad, Farzad
• Application Number:
  US11/563,835
• Publication Number:
  US 20080126100A1
• Time in Patent Office: 2,562 Days
• Field of Search: 704/240
• US Class Current: 704/275
• CPC Class Codes: G10L 15/08 Speech classification or search
• CPC Class Codes: G10L 15/20 Speech recognition techniques specially adapted for robustness in adverse environments, e.g. in noise, of stress induced speech G10L21/02 takes precedence
• CPC Class Codes: G10L 15/26 Speech to text systems G10L15/08 takes precedence

Abstract

A speech recognition method includes the steps of receiving input speech containing vocabulary, processing the input speech with a grammar to obtain N-best hypotheses and associated parameter values, and determining whether a first-best hypothesis of the N-best hypotheses is confusable with any vocabulary within the grammar. The first-best hypothesis is accepted as recognized speech corresponding to the received input speech if the first-best hypothesis is not determined to be confusable with any vocabulary within the grammar. Where the first-best hypothesis is determined to be confusable, at least one parameter value of the first-best hypothesis can be compared to at least one threshold value, and accepting the second-best as the recognized speech, if its confidence score is within certain lower and upper threshold values and is not confusable with the first-best. The first-best hypothesis can be accepted as recognized speech corresponding to the received input speech, if the parameter value of the first-best hypothesis is greater than the threshold value.

Claims

1. A speech recognition method comprising the steps of:

(a) receiving input speech containing vocabulary via a microphone associated with an automatic speech recognition system;

(b) processing the input speech with a grammar to obtain N-best hypotheses and associated parameter values using at least one processor associated with the automatic speech recognition system;

(c) cross-referencing a first-best hypothesis of the N-best hypotheses against a list of known confusable vocabulary to determine whether the first-best hypothesis of the N-best hypotheses is confusable with any of the known confusable vocabulary;

(d) accepting the first-best hypothesis as recognized speech corresponding to the received input speech, if the first-best hypothesis is not determined to be confusable with any of the known confusable vocabulary;

(e) comparing at least one parameter value of the first-best hypothesis to at least one threshold value, if the first-best hypothesis is determined to be confusable with any of the known confusable vocabulary;

(f) accepting the first-best hypothesis as recognized speech corresponding to the received input speech, if the at least one parameter value of the first- best hypothesis is greater than the at least one threshold value;

(g) determining if a second-best hypothesis of the N-best hypotheses is confusable with the first-best hypothesis, if the at least one parameter value of the first-best hypothesis is not greater than the at least one threshold value;

and (h) accepting the second-best hypothesis as recognized speech corresponding to the received input speech, if the second-best hypothesis is determined to be confusable with the first-best hypothesis;

(h1) determining if a confidence score of the second-best hypothesis is between lower and upper threshold values;

and (i) accepting the second-best hypothesis as recognized speech corresponding to the received input speech, if the confidence score is determined to be within the lower and upper threshold values.

2. The method of claim 1, further comprising the steps of:

(i1) setting the first-best hypothesis as recognized speech corresponding to the received input speech, if the second-best hypothesis is not determined to be to be confusable with the first-best hypothesis.

3. The method of claim 2, further comprising the step of:

(i2) transmitting a pardon message after setting the first-best hypothesis as the recognized speech.

4. The method of claim 3, further comprising the steps of:

(i4) presenting the first-best hypothesis to a user for confirmation;

and (i5) accepting the first-best hypothesis as recognized speech corresponding to the received input speech, if the first-best hypothesis is confirmed by the user.

5. The method of claim 1, wherein the at least one threshold value includes a plurality of threshold values, each corresponding with an individual hypothesis that is confusable with the first-best hypothesis.

6. The method of claim 1, wherein the at least one threshold value includes threshold values for a given hypothesis that is confusable with the first-best hypothesis, wherein the threshold values vary depending on a grammar being used in the processing step.

7. The method of claim 1, wherein the at least one threshold value includes threshold values for a given hypothesis that is confusable with the first-best hypothesis, wherein the threshold values vary by user.

8. The method of claim 1, wherein the at least one parameter value is a confidence value.

9. A speech recognition method comprising the steps of:

(a) receiving input speech containing vocabulary via a microphone associated with an automatic speech recognition system;

(b) processing the input speech with a grammar to obtain N-best hypotheses and associated parameter values using at least one processor associated with the automatic speech recognition system;

(c) cross-referencing a first-best hypothesis of the N-best hypotheses against a list of known confusable vocabulary to determine whether the first-best hypothesis of the N-best hypotheses is confusable with any of the known confusable vocabulary;

(d) accepting the first-best hypothesis as recognized speech corresponding to the received input speech, if the first-best hypothesis is not determined to be confusable with any of the known confusable vocabulary;

(e) comparing at least one parameter value of the first-best hypothesis to at least one threshold value, if the first-best hypothesis is determined to be confusable with any of the known confusable vocabulary; and

(f) accepting the first-best hypothesis as recognized speech corresponding to the received input speech, if the at least one parameter value of the first-best hypothesis is greater than the at least one threshold value;

(g) determining if a second-best hypothesis of the N-best hypotheses is confusable with the first-best hypothesis, if the at least one parameter value of the first-best hypothesis is not greater than the at least one threshold value;

(h) determining if a confidence score of the second-best hypothesis is between lower and upper threshold values; and

(i) accepting the second-best hypothesis as recognized speech corresponding to the received input speech, if the confidence score is determined to be within the lower and upper threshold values.

10. The method of claim 9, further comprising the steps of:

(m1) setting the first-best hypothesis as recognized speech corresponding to the received input speech, if the confidence score is not determined to be within the lower and upper threshold values.

11. The method of claim 10, further comprising the step of:

(m2) transmitting a pardon message after setting the first-best hypothesis as recognized speech.

12. The method of claim 11, further comprising the steps of:

(m4) presenting the first-best hypothesis to a user for confirmation; and

(m5) accepting the first-best hypothesis as recognized speech corresponding to the received input speech, if the first-best hypothesis is confirmed by the user.