System and method for adaptive automatic error correction

1. A speech recognition system having adaptive automatic error and mismatch correction comprising:

a computer storage medium and a computer program code mechanism embedded in the computer storage medium for causing a computer to interpret a string of speech;

an automatic error and mismatch correction module stored on said computer program code mechanism for use with receiving said string of interpreted speech;

a plurality of automatic error and mismatch correction models stored on said computer program code mechanism in electronic communication with said automatic error and mismatch correction module operatively configured to detect errors and mismatches in said string of interpreted speech and generate a process report;

an automatic error correction model adaptation module operatively configured to improve said automatic error and mismatch correction models based on said process report; and

a classifier module stored on said computer program code mechanism in electronic communication with said automatic error and mismatch correction module operatively configured to correct errors and mismatches in said string of interpreted speech.

2. The device of claim 1 further comprising a postprocessor stored on said computer program code mechanism for use with replacing said errors and mismatch in said string of interpreted speech with corrected and modified words.

3. The device of claim 1 further comprising a correction editing client stored on said computer program code mechanism for generating a process report pairing said errors and mismatches in said string of interpreted speech with corrected and modified words.

4. The device of claim 3 further comprising an automatic error and mismatch correction model adaptation module for use with modifying said automatic error and mismatch correction mold based upon comparing interpreted speech and final edited documents to identify and select errors and mismatches.

5. A method for adaptive automatic error and mismatch correction in a speech recognition system having an automatic error and mismatch correction module, automatic error and mismatch correction models, and a classifier module, the method comprising the steps of:

receiving a string of interpreted speech into the automatic error and mismatch correction module;

identifying target tokens in said string of interpreted speech;

selecting an automatic error and mismatch correction model from a set comprising a specific user model, a specific site model, and a factory model;

creating a set of classifier features according to requirements of the automatic error and mismatch correction model;

comparing said target tokens against said classifier features to detect errors and mismatches and to classify said target tokens;

correcting said target tokens for which said errors and mismatches are detected;

correcting said string of interpreted speech by replacement of said erroneous target tokens with said corrected target tokens;

evaluating the performance of said classifiers and identifying the best performing classifiers; and

improving said error and mismatch correction models by implementing said best performing. classifiers.

6. The method of claim 5 wherein the automatic error and mismatch adaptation module generates and replaces the existing error and mismatch classifier data.

7. The method of claim 5 wherein the classifier features are created according to the requirements of an automatic error and mismatch correction model associated with the speech recognition system.

8. A method for adaptive automatic error and mismatch correction in a speech recognition system having an automatic error correction model adaptation module comprising the steps of:

receiving a string of interpreted speech in the automatic error correction model adaptation module;

receiving a corresponding string of the final, edited form of said interpreted speech;

comparing said string of interpreted speech and said corresponding string of the final, edited form of the interpreted speech;

identifying mismatching target tokens in said string of interpreted speech and their corresponding targets in said final edited form of the interpreted speech;

creating patterns or rules consisting of source patterns in said string of interpreted speech and re-write patterns in said final, edited form of the interpreted speech;

creating a set of classifier features based on one of a plurality of automatic error and mismatch correction models;

evaluating the performance of the patterns or rules by determining the effectiveness or success of classifiers in identifying correct and incorrect patterns in interpreted speech using metrics chosen from a group comprising recall, precision, and F-Measure;

using said metrics to select classifiers.

9. The method of claim 8 wherein the classifier features are created according. to the requirements of an automatic error and mismatch correction model associated with the speech recognition system.

10. The method of claim 9 further comprising the step of generating said automatic error and mismatch correction model based upon the comparison of the errors and/or mismatches in said string of interpreted speech and the final edited form of this interpreted speech.

11. The method of claim 10 wherein said automatic error and mismatch correction model is replaced based upon a process report.