Dialect-specific acoustic language modeling and speech recognition

• US 8,583,432 B1
• Filed: 07/25/2012
• Issued: 11/12/2013
• Est. Priority Date: 07/18/2012
• Status: Expired due to Fees

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Biadsy, Fadi, Mangu, Lidia, Soltau, Hagen
• Primary Examiner(s): Guerra-Erazo, Edgar
• Application Number:
  US13/557,983
• Time in Patent Office: 475 Days
• Field of Search: 704/235, 704/243, 704/245, 704/254, 704/9, 704/10
• US Class Current: 704/235
• CPC Class Codes: G10L 15/005 Language recognition
• CPC Class Codes: G10L 15/063 Training
• CPC Class Codes: G10L 15/18 using natural language modelling

Abstract

Methods and systems for automatic speech recognition and methods and systems for training acoustic language models are disclosed. One system for automatic speech recognition includes a dialect recognition unit and a controller. The dialect recognition unit is configured to analyze acoustic input data to identify portions of the acoustic input data that conform to a general language and to identify portions of the acoustic input data that conform to at least one dialect of the general language. In addition, the controller is configured to apply a general language model and at least one dialect language model to the input data to perform speech recognition by dynamically selecting between the models in accordance with each of the identified portions.

Claims

1. A system for training acoustic models for automatic speech recognition comprising:

a dialect recognition unit configured to distinguish phones of a standard form language and at least one dialect for the language and to identify portions of an acoustic training data set that conform to the at least one dialect based on the distinguished phones of the at least one dialect; and

a controller, implemented by at least one processor, configured to train at least one dialect language model for automatic speech recognition based on the portions of the acoustic training data set that are identified as conforming to the at least one dialect.

2. The system of claim 1, wherein the dialect recognition unit is further configured to identify portions of the acoustic training data set that conform to the standard form language.

3. The system of claim 2, wherein the controller is further configured to train a standard form language model for automatic speech recognition based on the portions of the acoustic training data set that are identified as conforming to the standard form language.

4. The system of claim 2, wherein the controller is further configured to train the at least one dialect language model based on the portions of the acoustic training data set that are identified as conforming to the standard form language.

5. The system of claim 4, wherein each of the portions is a respective utterance in the training data set.

6. The system of claim 5, wherein the dialect recognition unit is further configured to determine posterior probability scores for each of the utterances indicating a respective probability that the corresponding utterance conforms to the standard form language and a respective probability that the corresponding utterance conforms to the at least one dialect.

7. The system of claim 6, wherein the controller is further configured to weight each of the utterances by the corresponding posterior probability scores for the utterance.

8. The system of claim 1, wherein the dialect recognition unit is further configured to employ general phone representations for the standard form language and to employ phone representations for the at least one dialect that are constructed by adapting the general phone representations.

9. The system of claim 1, wherein the dialect recognition unit further comprises a support vector machine classifier that is based on kernel values for different pairs of dialects of the standard form language.

10. A system for automatic speech recognition comprising:

a dialect recognition unit configured to analyze acoustic input data to identify portions of the acoustic input data that conform to a standard form language and to identify portions of the acoustic input data that conform to at least one dialect of the standard form language; and

a controller, implemented by at least one hardware processor, configured to apply a standard form language model and at least one dialect language model to the input data to perform speech recognition by dynamically selecting between the models in accordance with each of the identified portions.

11. The system of claim 10, wherein the dialect recognition unit is further configured to apply a single decoding graph including different decision trees corresponding to the standard form language model and the at least one dialect language model.

12. The system of claim 10, wherein the dialect recognition unit is further configured to distinguish phones of the standard form language and the at least one dialect.

13. A system for automatic speech recognition comprising:

a dialect recognition unit configured to analyze acoustic input data to compute probabilities that portions of the input data conform to a standard form language and probabilities that the portions of the input data conform to at least one dialect of the standard form language; and

a controller, implemented by at least one processor, configured to apply a standard form language model and at least one dialect language model to the input data to perform speech recognition by weighting the models in accordance with each of the computed probabilities.

14. The system of claim 13, wherein the dialect recognition unit is further configured to apply a single decoding graph including different decision trees corresponding to the standard form language model and the at least one dialect language model.

15. The system of claim 13, wherein the dialect recognition unit is further configured to distinguish phones of the standard form language and the at least one dialect.

16. The system of claim 13, wherein each of the portions is a respective utterance in the input data.

17. The system of claim 16, wherein the dialect recognition unit is further configured to compute, for each of the utterances, a weighted sum of log likelihoods determined in accordance with the standard form language model and the at least one dialect language model.

18. The system of claim 13, wherein the dialect recognition unit is further configured to employ general phone representations for the standard form language and phone representations for the at least one dialect that are constructed by adapting the general phone representations.

19. The system of claim 18, wherein the dialect recognition unit is further configured to employ a support vector machine classifier that is based on kernel values for different pairs of dialects of the standard form language.

20. The system of claim 13, wherein the dialect recognition unit is further configured to analyze the acoustic input data to compute the respective probabilities in a single decoding pass.