Granted Patent B2

US 9,431,011 · App. 14/488,844 · Granted Aug 30, 2016

Abstract

Claims (53)

1. A method comprising:

receiving user speech;

identifying a user dialect in the user speech;

selecting a set of phoneme alternatives representing the user dialect from a library of alternative pronunciation models, each alternative generated by the process of:

(i) identifying a generic model of speech composed of phonemes;

(ii) identifying a family of interchangeable phonemic alternatives for a phoneme in the generic model of speech;

(iii) labeling each interchangeable phonemic alternative in the family as referring to the phoneme; and

(iv) generating a pronunciation model which substitutes the family of interchangeable phonemic alternatives for the phoneme; and

recognizing, via a processor, the user speech using the selected set of phoneme alternatives in the pronunciation model.

2. The method of claim 1 , wherein identifying the user dialect further comprises:

recognizing the user speech with a plurality of dialect models;

eliminating dialect models in the plurality of dialect models which do not phonemically match the user speech until a single dialect model remains; and

identifying the remaining single dialect model as the user dialect model.

3. The method of claim 1 , wherein the generic model of speech is a vocal tract length normalized acoustic model.

4. The method of claim 3 , wherein the generic model does not include non-dialectal acoustic variation.

5. The method of claim 1 , wherein interchangeable phonemic alternatives represent a same phoneme for different dialectal classes.

6. The method of claim 1 , wherein an interchangeable phonemic alternative comprises a string of phonemes.

7. The method of claim 1 , further comprising repeating the receiving, the identifying, the selecting, and the recognizing for each speaker in a group of speakers, to yield a different model for each speaker.

8. A system comprising:

a processor; and

a computer-readable storage medium having instructions stored which, when executed by the processor, cause the processor to perform operations comprising:

receiving user speech;

identifying a user dialect in the user speech;

selecting a set of phoneme alternatives representing the user dialect from a library of alternative pronunciation models, each alternative generated by the process of:

(i) identifying a generic model of speech composed of phonemes;

(ii) identifying a family of interchangeable phonemic alternatives for a phoneme in the generic model of speech;

(iii) labeling each interchangeable phonemic alternative in the family as referring to the phoneme; and

(iv) generating a pronunciation model which substitutes the family of interchangeable phonemic alternatives for the phoneme; and

recognizing the user speech using the selected set of phoneme alternatives in the pronunciation model.

9. The system of claim 8 , wherein identifying the user dialect further comprises:

recognizing the user speech with a plurality of dialect models;

eliminating dialect models in the plurality of dialect models which do not phonemically match the user speech until a single dialect model remains; and

identifying the remaining single dialect model as the user dialect model.

10. The system of claim 8 , wherein the generic model of speech is a vocal tract length normalized acoustic model.

11. The system of claim 10 , wherein the generic model does not include non-dialectal acoustic variation.

12. The system of claim 8 , wherein interchangeable phonemic alternatives represent a same phoneme for different dialectal classes.

13. The system of claim 8 , wherein an interchangeable phonemic alternative comprises a string of phonemes.

14. The system of claim 8 , the computer-readable storage medium having additional instructions stored which, when executed by the processor, result in operations comprising repeating the receiving, the identifying, the selecting, and the recognizing for each speaker in a group of speakers, to yield a different pronunciation model for each speaker.

15. A computer-readable storage device having instructions stored which, when executed by a computing device, cause the computing device to perform operations comprising:

receiving user speech;

identifying a user dialect in the user speech;

selecting a set of phoneme alternatives representing the user dialect from a library of alternative pronunciation models, each alternative generated by the process of:

(i) identifying a generic model of speech composed of phonemes;

(ii) identifying a family of interchangeable phonemic alternatives for a phoneme in the generic model of speech;

(iii) labeling each interchangeable phonemic alternative in the family as referring to the phoneme; and

(iv) generating a pronunciation model which substitutes the family of interchangeable phonemic alternatives for the phoneme; and

recognizing the user speech using the selected set of phoneme alternatives in the pronunciation model.

16. The computer-readable storage device of claim 15 , wherein identifying the user dialect further comprises:

recognizing the user speech with a plurality of dialect models;

eliminating dialect models in the plurality of dialect models which do not phonemically match the user speech until a single dialect model remains; and

identifying the remaining single dialect model as the user dialect model.

17. The computer-readable storage device of claim 15 , wherein the generic model of speech is a vocal tract length normalized acoustic model.

18. The computer-readable storage device of claim 17 , wherein the generic model does not include non-dialectal acoustic variation.

Assignments (15)