Specific task composite acoustic models

• US 6,260,014 B1
• Filed: 09/14/1998
• Issued: 07/10/2001
• Est. Priority Date: 09/14/1998
• Status: Expired due to Term

• Current Assignee: Nuance Communications, Inc. (Microsoft Corporation)
• Original Assignee: International Business Machines Corporation
• Inventors: Bahl, Lalit Rai, Lubensky, David, Roukos, Salim, Padmanabhan, Mukund
• Primary Examiner(s): Tsang, Fan
• Assistant Examiner(s): Opsasnick, Michael N.
• Application Number:
  US09/153,222
• Time in Patent Office: 1,030 Days
• Field of Search: 704/255, 704/231, 704/254
• US Class Current: 704/254
• CPC Class Codes: G10L 15/063 Training
• CPC Class Codes: G10L 15/183 using context dependencies, e.g. language models
• CPC Class Codes: G10L 15/187 Phonemic context, e.g. pronunciation rules, phonotactical constraints or phoneme n-grams
• CPC Class Codes: G10L 15/197 Probabilistic grammars, e.g. word n-grams
• CPC Class Codes: G10L 15/22 Procedures used during a speech recognition process, e.g. man-machine dialogue
• CPC Class Codes: G10L 15/26 Speech to text systems G10L15/08 takes precedence
• CPC Class Codes: G10L 2015/223 Execution procedure of a spoken command
• CPC Class Codes: G10L 2015/228 of application context

Abstract

A method for recognizing speech includes the steps of providing a generic model having a baseform representation of a vocabulary of words, identifying a subset of words relating to an application, constructing a task specific model for the subset of words, constructing a composite model by combining the generic and task specific models and modifying the baseform representation of the subset of words such that the subset of words are recognized by the task specific model. A system for recognizing speech includes a composite model having a generic model having a generic baseform representation of a vocabulary of words and a task specific model for recognizing a subset of words relating to an application wherein the subset of words are recognized using a modified baseform representation. A recognizer compares words input thereto with the generic model for words other than the subset of words and with the task specific model for the subset of words.

Claims

1. A method for recognizing speech comprising the steps of:

providing a generic model having a generic phonetic baseform representation of a vocabulary of words;

identifying a subset of words relating to an application;

constructing a task specific model for the subset of words;

constructing a composite model by combining the generic model and the task specific model; and

modifying the generic phonetic baseform representation of the subset of words such that the task specific model is used to recognize the subset of words.

2. The method as recited in claim 1, wherein the step of identifying a subset of words includes the step of identifying a subset words pertinent to a particular task.

3. The method as recited in claim 1, wherein the step of constructing a task specific model includes the step of constructing the task specific model by utilizing a mixture of gaussian functions to model a probability density function of feature vectors for different words in the vocabulary.

4. The method as recited in claim 1, wherein the step of constructing a task specific model includes the step of constructing the task specific model by utilizing a mixture of non-gaussian functions to model a probability density function of feature vectors for different words in the vocabulary.

5. The method as recited in claim 1, wherein the step of constructing a task specific model includes the step of constructing the task specific model by utilizing a neural network to model a probability density function of feature vectors for different words in the vocabulary.

6. The method as recited in claim 1, further comprising the step of estimating parameters of speech using an estimation technique.

7. The method as recited in claim 1, wherein an estimation technique for the generic model is different from an estimation technique for the task specific model.

8. The method as recited in claim 1, wherein the step of constructing a task specific model includes the step of constructing the task specific model by utilizing a different construction technique than a construction technique used for the generic model.

9. The method as recited in claim 1, wherein generic model and the task specific model include parametric models and further comprising the step of modeling a probability density function of feature vectors for task specific words and generic words.

10. The method as recited in claim 9, wherein the generic model and the task specific model have different probability density functions.

11. The method as recited in claim 1, further comprising the step of interchanging the task specific model with a different task specific model to create a new composite model.

12. A method for recognizing speech comprising the steps of:

constructing a generic model having a phonetic baseform representation of a vocabulary of words;

constructing a task specific model for a subset of words;

constructing a composite model by combining the generic model and the task specific model;

modifying the phonetic baseform representation of the subset of words such that the subset of words are recognized by the task specific model; and

recognizing words other than the task specific words using the generic model.

13. The method as recited in claim 12, wherein the step of constructing a task specific model includes the step of constructing the task specific model by utilizing a mixture of non-gaussian functions to model a probability density function of feature vectors for different words in the vocabulary.

14. The method as recited in claim 12, wherein the step of constructing a task specific model includes the step of constructing the task specific model by utilizing a mixture of gaussian functions to model a probability density function of feature vectors for different words in the vocabulary.

15. The method as recited in claim 12, wherein the step of constructing a task specific model includes the step of constructing the task specific model by utilizing a neural network to model a probability density function of feature vectors for different words in the vocabulary.

16. The method as recited in claim 12, further comprising the step of estimating parameters of speech using an estimation technique.

17. method as recited in claim 12, wherein an estimation technique for the generic model is different from an estimation technique for the task specific model.

18. The method as recited in claim 12, wherein the step of constructing a task specific model includes the step of constructing the task specific model by utilizing a different construction technique than a construction technique used for the generic model.

19. The method as recited in claim 12, wherein generic model and the task specific model include parametric models and further comprising the step of modeling a probability density function of feature vectors for task specific words and generic words.

20. The method as recited in claim 19, wherein the generic model and the task specific model have different probability density functions.

21. The method as recited in claim 12, further comprising the step of interchanging the task specific model with a different task specific model to create a new composite model.

27. The system as recited in claim 20, further comprising a plurality of task specific models for applying to a plurality of applications.

22. A system for recognizing speech comprising:

a composite model including;

a generic model having a baseform representation of a vocabulary of words; and

a task specific model for recognizing a subset of words relating to an application wherein the subset of words are recognized using a modified baseform representation; and

a recognizer for comparing words input thereto with the generic model for words other than the subset of words and with the task specific model for the subset of words.

23. The system as recited in claim 22, wherein the recognizer includes a processor.

24. The system as recited in claim 22, wherein the generic model and the task specific model use different parametric models for modeling probability density functions.

25. The system as recited in claim 22, wherein the generic model and the task specific model are constructed using different construction techniques.

26. The system as recited in claim 22, wherein the generic model and the task specific model provide different probability estimation techniques for recognizing speech.