SYSTEM AND METHOD OF WORD LATTICE AUGMENTATION USING A PRE/POST VOCALIC CONSONANT DISTINCTION

• US 20090112591A1
• Filed: 10/31/2007
• Published: 04/30/2009
• Est. Priority Date: 10/31/2007
• Status: Active Grant

• Current Assignee: Nuance Communications, Inc. (Microsoft Corporation)
• Original Assignee: AT&T Labs Incorporated (AT&T, Inc.)
• Inventors: KIM, Yeon-Jun, Conkie, Alistair, Syrdal, Ann K., Ljolje, Andrej
• Granted Patent: US 8,024,191 B2
• Time in Patent Office: Days
• US Class Current: 704/246
• CPC Class Codes: G10L 15/02 Feature extraction for speech recognition; Selection of recognition unit
• CPC Class Codes: G10L 25/78 Detection of presence or absence of voice signals switching of direction of transmission by voice frequency in two-way loud-speaking telephone systems H04M9/10

Abstract

Disclosed are systems and methods for recognizing speech in a spoken dialogue system. The method includes (1) receiving an input speech having at least one pre-vocalic consonant or at least one post-vocalic consonant, (2) generating at least one output lattice that calculates a first score by comparing the input speech to a training model to provide a result; (3) distinguishing between the at least one pre-vocalic consonant and the at least one post-vocalic consonant in the input speech, (4) calculating a second score by measuring a similarity between the at least one pre-vocalic consonant or the at least one post vocalic consonant in the input speech and the first score, (5) determining at least one category for at least one pre-vocalic match or mismatch or at least one post-vocalic match or mismatch by using the second score, and (6) refining the results of the an automated speech recognition (ASR) system by using the at least one category for at least one pre-vocalic match or mismatch or at least one post-vocalic match or mismatch.

Claims

1. The method for recognizing speech, the method comprising:

receiving an input speech having at least one pre-vocalic consonant or at least one post-vocalic consonant;

generating at least one output lattice that calculates a first score by comparing the input speech to a training model to provide a result;

distinguishing between the at least one pre-vocalic consonant and the at least one post-vocalic consonant in the input speech;

calculating a second score by measuring a similarity between the at least one pre-vocalic consonant or the at least one post-vocalic consonant in the input speech and the first score;

determining at least one category for at least one pre-vocalic match or mismatch or at least one post-vocalic match or mismatch by using the second score; and

refining the results of the an automated speech recognition (ASR) system by using the at least one category for at least one pre-vocalic match or mismatch or at least one post-vocalic match or mismatch.

2. The method of claim 1, wherein if there is a match between the second output lattice and the second score, a word probability is increased.

3. The method of claim 1, wherein at least one output lattice comprises syllabified words.

4. The method of claim 1, wherein if there is a mismatch between the second output lattice and the second score, a word probability is decreased.

5. The method of claim 1, wherein a goodness score is determined by calculating the similarity between the input speech and the training model.

6. The method of claim 1, wherein the ASR system is trained to distinguish each of the pre-vocalic consonants and the post-vocalic consonants.

7. The method of claim 1, wherein the training model distinguishes between the at least one pre-vocalic consonant and the at least one post-vocalic consonant can be accomplished by using hidden Markov models (HMMs), support vector machines (SVMs) and neural networks (NNs).

9. The system of claim 1, wherein if there is a match between the second output lattice and the second score, a word probability is increased.

10. The system of claim 1, wherein at least one output lattice comprises syllabified words.

11. The system of claim 1, wherein if there is a mismatch between the second output lattice and the second score, a word probability is decreased.

12. The system of claim 1, wherein a goodness score is determined by calculating the similarity between the input speech and the training model.

13. The system of claim 1, wherein the ASR system is trained to distinguish each of the pre-vocalic consonants and the post-vocalic consonants.

14. The system of claim 1, wherein the training model distinguishes between the at least one pre-vocalic consonant and the at least one post-vocalic consonant can be accomplished by using hidden Markov models (HMMs), support vector machines (SVMs) and neural networks (NNs).

16. The computer-readable medium of claim 1, wherein if there is a match between the second output lattice and the second score, a word probability is increased.

17. The computer-readable medium of claim 1, wherein at least one output lattice comprises syllabified words.

18. The computer-readable medium of claim 1, wherein if there is a mismatch between the second output lattice and the second score, a word probability is decreased.

19. The computer-readable medium of claim 1, wherein a goodness score is determined by calculating the similarity between the input speech and the training model.

20. The computer-readable medium of claim 1, wherein the ASR system is trained to distinguish each of the pre-vocalic consonants and the post-vocalic consonants.

21. The computer-readable medium of claim 1, wherein the training model distinguishes between the at least one pre-vocalic consonant and the at least one post-vocalic consonant can be accomplished by using hidden Markov models (HMMs), support vector machines (SVMs) and neural networks (NNs).

8. A system for recognizing speech, the system comprising:

a module configured to receive an input speech having at least one pre-vocalic consonant or at least one post-vocalic consonant;

a module configured to generate at least one output lattice that calculates a first score by comparing the input speech to a training model to provide a result;

a module configured to distinguish between the at least one pre-vocalic consonant and the at least one post-vocalic consonant in the input speech;

a module configured to calculate a second score by measuring a similarity between the at least one pre-vocalic consonant or the at least one post vocalic consonant in the input speech and the first score;

a module configured to determine at least one category for at least one pre-vocalic match or mismatch or at least one post-vocalic match or mismatch by using the second score; and

a module configured to refine the results of the an automated speech recognition (ASR) system by using the at least one category for at least one pre-vocalic match or mismatch or at least one post-vocalic match or mismatch.

15. A computer-readable medium storing instructions for controlling a computing device to process speech, the instructions comprising:

receiving an input speech having at least one pre-vocalic consonant or at least one post-vocalic consonant;

generating at least one output lattice that calculates a first score by comparing the input speech to a training model to provide a result;

distinguishing between the at least one pre-vocalic consonant and the at least one post-vocalic consonant in the input speech;

calculating a second score by measuring a similarity between the at least one pre-vocalic consonant or the at least one post-vocalic consonant in the input speech and the first score;

determining at least one category for at least one pre-vocalic match or mismatch or at least one post-vocalic match or mismatch by using the second score; and

refining the results of the an automated speech recognition (ASR) system by using the at least one category for at least one pre-vocalic match or mismatch or at least one post-vocalic match or mismatch.