SYSTEM AND METHOD FOR DISCRIMINATIVE PRONUNCIATION MODELING FOR VOICE SEARCH

• US 20100125457A1
• Filed: 11/19/2008
• Published: 05/20/2010
• Est. Priority Date: 11/19/2008
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: AT&T Intellectual Property I LP (AT&T, Inc.)
• Inventors: Gilbert, Mazin, Ljolje, Andrej, Conkie, Alistair D.
• Granted Patent: US 8,296,141 B2
• Time in Patent Office: Days
• US Class Current: 704/254
• CPC Class Codes: G10L 15/063 Training
• CPC Class Codes: G10L 2015/025 Phonemes, fenemes or fenones being the recognition units

Abstract

Disclosed herein are systems, computer-implemented methods, and computer-readable media for speech recognition. The method includes receiving speech utterances, assigning a pronunciation weight to each unit of speech in the speech utterances, each respective pronunciation weight being normalized at a unit of speech level to sum to 1, for each received speech utterance, optimizing the pronunciation weight by (1) identifying word and phone alignments and corresponding likelihood scores, and (2) discriminatively adapting the pronunciation weight to minimize classification errors, and recognizing additional received speech utterances using the optimized pronunciation weights. A unit of speech can be a sentence, a word, a context-dependent phone, a context-independent phone, or a syllable. The method can further include discriminatively adapting pronunciation weights based on an objective function. The objective function can be maximum mutual information (MMI), maximum likelihood (MLE) training, minimum classification error (MCE) training, or other functions known to those of skill in the art. Speech utterances can be names. The speech utterances can be received as part of a multimodal search or input. The step of discriminatively adapting pronunciation weights can further include stochastically modeling pronunciations.

Claims

1. A computer-implemented method of performing speech recognition, the method comprising:

receiving speech utterances;

assigning a pronunciation weight to each unit of speech in the speech utterances, each respective pronunciation weight being normalized at a unit of speech level to sum to 1;

for each received speech utterance, optimizing the pronunciation weight by;

(1) identifying word and phone alignments and corresponding likelihood scores;

(2) discriminatively adapting the pronunciation weight to minimize classification errors; and

recognizing additional received speech utterances using the optimized pronunciation weights.

2. The computer-implemented method of claim 1, wherein a unit of speech is one of a sentence, a word, a context-dependent phone, a context-independent phone, and a syllable.

3. The computer-implemented method of claim 1, wherein discriminatively adapting pronunciation weights is based on an objective function.

4. The computer-implemented method of claim 3, wherein the objective function is one of maximum mutual information (MMI), maximum likelihood (MLE) training, and minimum classification error (MCE) training.

5. The computer-implemented method of claim 1, wherein the speech utterances include names.

6. The computer-implemented method of claim 1, wherein the speech utterances are part of multimodal input.

7. The computer-implemented method of claim 1, wherein discriminatively adapting pronunciation weights further comprises stochastically modeling pronunciations.

18. The computer-implemented method of claim 7, wherein the objective function is one of maximum mutual information (MMI), maximum likelihood (MLE) training, and minimum classification error (MCE) training.

8. A system for performing speech recognition, the system comprising:

a module configured to receive speech utterances;

a module configured to assign a pronunciation weight to each unit of speech in the speech utterances, each respective pronunciation weight being normalized at a unit of speech level to sum to 1;

a module configured to optimize the pronunciation weight for each received speech utterance by;

(1) identifying word and phone alignments and corresponding likelihood scores;

(2) discriminatively adapting the pronunciation weight to minimize classification errors; and

a module configured to recognize additional received speech utterances using the optimized pronunciation weights.

9. The system of claim 8, wherein a unit of speech is one of a sentence, a word, a context-dependent phone, a context-independent phone, and a syllable.

10. The system of claim 8, wherein discriminatively adapting pronunciation weights is based on an objective function.

11. The system of claim 10, wherein the objective function is one of maximum mutual information (MMI), maximum likelihood (MLE) training, and minimum classification error (MCE) training.

12. The system of claim 8, wherein the speech utterances include names.

13. The system of claim 8, wherein the speech utterances are part of multimodal input.

14. The system of claim 8, wherein discriminatively adapting pronunciation weights further comprises stochastically modeling pronunciations.

15. A tangible computer-readable medium storing a computer program having instructions for performing speech recognition, the instructions comprising:

receiving speech utterances;

assigning a pronunciation weight to each unit of speech in the speech utterances, each respective pronunciation weight being normalized at a unit of speech level to sum to 1;

for each received speech utterance, optimizing the pronunciation weight by;

(1) identifying word and phone alignments and corresponding likelihood scores;

(2) discriminatively adapting the pronunciation weight to minimize classification errors; and

recognizing additional received speech utterances using the optimized pronunciation weights.

16. The computer-implemented method of claim 15, wherein a unit of speech is one of a sentence, a word, a context-dependent phone, a context-independent phone, and a syllable.

17. The computer-implemented method of claim 15, wherein discriminatively adapting pronunciation weights is based on an objective function.

19. The computer-implemented method of claim 15, wherein the speech utterances include names.

20. The computer-implemented method of claim 15, wherein the speech utterances are part of multimodal input.