SYSTEMS AND METHODS OF IMPROVING AUTOMATED SPEECH RECOGNITION ACCURACY USING STATISTICAL ANALYSIS OF SEARCH TERMS

• US 20090292538A1
• Filed: 05/20/2008
• Published: 11/26/2009
• Est. Priority Date: 05/20/2008
• Status: Active Grant

• Current Assignee: Calabrio
• Original Assignee: Calabrio
• Inventors: Barnish, David M.
• Granted Patent: US 8,543,393 B2
• Time in Patent Office: Days
• US Class Current: 704/232
• CPC Class Codes: G10L 15/01 Assessment or evaluation of speech recognition systems
• CPC Class Codes: G10L 15/05 Word boundary detection
• CPC Class Codes: G10L 2015/025 Phonemes, fenemes or fenones being the recognition units

Abstract

Systems and methods of improving speech recognition accuracy using statistical analysis of word or phrase-based search terms are disclosed. An illustrative system for statistically analyzing search terms includes an interface adapted to receive a text-based search term, a textual-linguistic analysis module that detects textual features within the search term and generates a first score, a phonetic conversion module that converts the search term into a phoneme string, a phonetic-linguistic analysis module that detects phonemic features within the phoneme string and generates a second score, and a score normalization module that normalizes the first and second scores and outputs a search term score to a user or process.

Claims

1. A system of statistically analyzing search terms for detection by a speech recognition engine, comprising

an interface adapted to receive a text-based search term;

a textual-linguistic analysis module adapted to detect one or more textual features within the search term and generate a first score associated with the search term;

a phonetic conversion module adapted convert the text-based search term into a phoneme string;

a phonemic-linguistic analysis module adapted to detect one or more phonemic features within the phoneme string and generate a second score associated with the search term;

a score normalization module adapted to normalize the first and second scores and output a search term score to a user or process.

2. The system of claim 1, wherein the phonetic conversion module includes a language model dictionary and a phonetic alphabet.

3. The system of claim 2, further including a phonetic generation algorithm or routine adapted to generate a phoneme string for phonemes not contained in the language model dictionary.

4. The system of claim 1, wherein the textual-linguistic analysis module includes a table of textual features and associated weighting factors.

5. The system of claim 4, wherein the first score is an unbounded score representing the sum of all textual features in the search term each adjusted by a corresponding weighting factor.

6. The system of claim 1, wherein the phonemic-linguistic analysis module includes a table of phonemic features and associated weighting factors.

7. The system of claim 6, wherein the second score is an unbounded score representing the sum of all phonemic features in the phoneme string each adjusted by a corresponding weighting factor.

8. The system of claim 1, wherein the search term score is a whole number score.

9. The system of claim 1, wherein the search term score corresponds to a probability that the search term will be accurately detected by a speech recognition engine.

10. The system of claim 1, wherein the system is a computer-assisted system.

11. A method of statistically analyzing search terms for detection by a speech recognition engine, comprising:

receiving a text-based search term from a user or process;

analyzing one or more textual features within the search term using a textual-linguistic analysis module;

computing a first score associated with the one or more textual features within the search term;

converting the search term into a phoneme string and analyzing one or more phonemic features within the phoneme string using a phonemic-linguistic analysis module;

computing a second score associated with the one or more phonemic features within the phoneme string; and

normalizing the first and second scores and outputting a search term score to the user or process.

12. The method of claim 11, wherein computing a first score associated with the one or more textual features within the search term includes applying weighting factors to each textual feature.

13. The method of claim 12, wherein the first score correlates to the likelihood that the text of the search term will be accurately detected by a speech recognition engine.

14. The method of claim 11, wherein computing a second score associated with the one or more phonemic features within the phoneme string includes applying weighting factors to each phonemic feature.

15. The method of claim 12, wherein the second score correlates to the likelihood that the phoneme string will be accurately detected by a speech recognition engine.

16. The method of claim 11, wherein converting the search term into a phoneme string is accomplished using a phonetic conversion module and a language model dictionary.

17. The method of claim 11, wherein normalizing the first and second scores includes applying normalization constants to the first and second scores.

18. The method of claim 11, wherein the method is a computer-assisted method executable on a computer-readable medium.

19. A method of statistically analyzing search terms for detection by a speech recognition engine, comprising:

providing a text-based search term to a search-term analysis module including a textual-linguistic analysis module and a phonemic-linguistic analysis module;

analyzing a search term and outputting a search term score correlating to a probability that the search term will be accurately detected by a speech recognition engine;

comparing the search term score against a threshold score; and

outputting the search term to a database if the search term score is at or above the threshold score.

20. The method of claim 19, wherein analyzing the search term and outputting a search term score includes:

performing a textual-linguistic analysis of the search term and generating a first score based on one or more textual features found in the search term;

converting the search term into a phoneme string;

performing a phonemic-linguistic analysis of the search term and generating a second score based on one or more phonemic features found in the phoneme string; and

normalizing the first and second scores to obtain a search term score.