Phonetic features for speech recognition

1. A method, comprising:

obtaining a first dictionary and a training data set associated with a speech recognition system;

computing one or more support parameters from the training data set;

transforming the first dictionary into a second dictionary, wherein the second dictionary is a function of one or more phonetic labels of the first dictionary;

using the one or more support parameters to select one or more samples from the second dictionary to create a set of one or more exemplar-based class identification features for a pattern recognition task; and

using the set of one or more exemplar-based class identification features to perform a pattern recognition task, said step comprising defining a sensitivity matrix to contain one or more exemplars from the training data set;

wherein the steps are carried out by a computer device.

2. The method of claim 1 , wherein creating a set of one or more exemplar-based class identification features comprises solving a sparse representation problem.

3. The method of claim 2 , wherein the sparse representation problem comprises representing a test feature vector as a functional transformation of one or more training examples and learning a set of coefficients for the functional transformation.

4. The method of claim 3 , wherein creating the set of one or more exemplar-based class identification features comprises weighting the set of one or more coefficients in a transformed sensitivity matrix.

5. The method of claim 1 , wherein using the set of one or more exemplar-based class identification features to perform a pattern recognition task comprises using the set of one or more exemplar-based class identification features for at least one of speech recognition, text classification and speaker recognition.

6. The method of claim 1 , wherein the sensitivity matrix comprises one or more training examples.

7. The method of claim 1 , further comprising transforming the sensitivity matrix into a second matrix wherein each sub-matrix in the second matrix contains class identity information about an exemplar from a same sub-matrix of the sensitivity matrix.

8. The method of claim 7 , wherein the class identity for each sub-matrix in the second matrix is created from a set of one or more indexes corresponding to a class identity in a same sub-matrix in the sensitivity matrix.

9. The method of claim 8 , wherein the one or more indexes corresponding to a class identity in a same sub-matrix correspond to one or more class labels.

10. The method of claim 9 , wherein the one or more class labels comprise at least one of context-independent phonetic labels, context-dependent phonetic labels, broad phonetic classes, speaker identify, and document classes.

11. The method of claim 7 , wherein a sub-matrix can be a vector or a group of one or more vectors.

12. The method of claim 1 , further comprising providing a system, wherein the system comprises one or more distinct software modules, each of the one or more distinct software modules being embodied on a tangible computer-readable recordable storage medium, and wherein the one or more distinct software modules comprise a sparse representation training engine module, a feature extractor module, a speech recognition engine module, and an acoustic model storage module executing on a hardware processor.

13. A system, comprising:

a memory; and

a processor operatively coupled to the memory and configured to:

obtain a first dictionary and a training data set associated with a speech recognition system;

compute one or more support parameters from the training data set;

transform the first dictionary into a second dictionary, wherein the second dictionary is a function of one or more phonetic labels of the first dictionary; and

use the one or more support parameters to select one or more samples from the second dictionary to create a set of one or more exemplar-based class identification features for a pattern recognition task; and

use the set of one or more exemplar-based class identification features to perform a pattern recognition task, said step comprising defining a sensitivity matrix to contain one or more exemplars from the training data set.

14. The system of claim 13 , wherein the creating a set of one or more exemplar-based class identification features step further comprises solving a sparse representation problem.

15. The system of claim 14 , wherein the sparse representation problem comprises representing a test feature vector as a functional transformation of one or more training examples and learning a set of coefficients for the functional transformation.

16. The system of claim 13 , wherein the creating the set of one or more exemplar-based class identification features step further comprises weighting the set of one or more coefficients in a transformed sensitivity matrix.

17. The system of claim 13 , wherein the using the set of one or more exemplar-based class identification features to perform a pattern recognition task step further comprises using the set of one or more exemplar-based class identification features for at least one of speech recognition, text classification and speaker recognition.

18. The system of claim 13 , wherein the processor operatively coupled to the memory is further configured to transform the sensitivity matrix into a second matrix wherein each sub-matrix in the second matrix contains class identity information about an exemplar from a same sub-matrix of the sensitivity matrix.

19. The system of claim 18 , wherein the class identity for each sub-matrix in the second matrix is created from a set of one or more indexes corresponding to a class identity in a same sub-matrix in the sensitivity matrix.

20. The system of claim 19 , wherein the one or more indexes corresponding to a class identity in a same sub-matrix correspond to one or more class labels.

21. An article of manufacture, comprising a computer readable storage medium having tangibly embodied thereon computer readable program code which, when executed, causes a processor device to:

obtain a first dictionary and a training data set associated with a speech recognition system;

compute one or more support parameters from the training data set;

transform the first dictionary into a second dictionary, wherein the second dictionary is a function of one or more phonetic labels of the first dictionary;

use the one or more support parameters to select one or more samples from the second dictionary to create a set of one or more exemplar-based class identification features for a pattern recognition task; and

use the set of one or more exemplar-based class identification features to perform a pattern recognition task, said step comprising defining a sensitivity matrix to contain one or more exemplars from the training data set.