Method for system combination in an audio analytics application

1. A method of system combination in an audio analytics application, comprising:

providing a plurality of language identification systems, wherein each of the language identification systems includes a plurality of probabilities, wherein each probability is associated with the system's ability to detect a particular language;

receiving data at the language identification systems, wherein the received data is different from data used to train the language identification systems;

determining a confidence measure for each of the language identification systems, wherein the confidence measure identifies a level of accuracy for each of the language identification systems at predicting a presence of the particular language in the received data based on an inverse entropy of a posterior probability distribution of each of the language identification systems, wherein the inverse entropy of the posterior probability of each of the language identification systems is based on a number of languages each of the language identification systems can identify and the plurality of probabilities of each of the language identification systems;

rank ordering each of the language identification systems based on the confidence measure; and

combining a subset of the language identification systems having confidence measures above a predetermined threshold.

2. The method of claim 1 , wherein the language identification systems have different feature extraction methods from each other.

3. The method of claim 1 , wherein the language identification systems have different modeling schemes from each other.

4. The method of claim 1 , wherein the language identification systems have different noise removal schemes from each other.

5. The method of claim 1 , wherein the received data and the data used to train the language identification systems include speech.

6. The method of claim 1 , wherein the received data includes an utterance.

7. The method of claim 1 , wherein determining a confidence measure for each of the language identification systems includes normalizing the inverse entropy value for each of the language identification systems.

8. The method of claim 1 , wherein the steps of determining the confidence measure and combining the language identification systems are repeated for each utterance in the received data.

9. The method of claim 1 , wherein the received data has different characteristics than the data used to train the language identification systems.

10. The method of claim 1 , further comprising using the confidence measure to identify which language its system is best at detecting in the received data.

11. The method of claim 1 , further comprising applying the combination of confidence measures to the received data to increase a performance metric of the language identification systems.

12. The method of claim 11 , wherein the performance metric is indicative of accuracy in detecting language.

13. A method of system combination in an audio analytics application, comprising:

providing a plurality of language identification systems, wherein each of the language identification systems includes a plurality of probabilities, wherein each probability is associated with the system's ability to detect a particular language;

receiving data at the language identification systems, wherein the received data has different characteristics than data used to train the language identification systems;

determining a confidence measure for each of the language identification systems using a portion of the received data, wherein the confidence measure identifies a level of accuracy for each of the language identification systems at predicting a presence of the particular language in the received data based on an inverse entropy of a posterior probability distribution of each of the language identification systems, wherein the inverse entropy of the posterior probability of each of the language identification systems is based on a number of languages each of the language identification systems can identify and the plurality of probabilities of each of the language identification systems; and

combining at least two of the language identification systems having confidence measures above a predetermined threshold.

14. The method of claim 13 , wherein the portion of the received data used to determine the confidence measure is less than 10% of the received data.

15. The method of claim 13 , wherein less than all of the language identification systems are combined according to the confidence measures.

16. The method of claim 13 , further comprising, before combining the at least two language identification systems, pruning the language identification systems based on their confidence measures.

17. A method of system combination in an audio analytics application, comprising:

providing a plurality of language identification systems trained on first data;

receiving a second data at the language identification systems, wherein the second data is different from the first data;

determining a confidence measure for each of the language identification systems, wherein the confidence measure identifies a level of accuracy for each of the language identification systems at predicting a presence of a particular language in the received data based on an inverse entropy of a posterior probability distribution of each of the language identification systems, wherein the inverse entropy of the posterior probability of each of the language identification systems is based on a number of languages each of the language identification systems can identify and a plurality of probabilities of each of the language identification systems;

combining a subset of the language identification systems having confidence measures above a predetermined threshold;

inputting third data different from the first and second data to the combination of language identification systems; and

identifying a language of the third data.

18. The method of claim 17 , wherein each of the language identification systems includes the plurality of probabilities, wherein each probability is associated with the system's ability to detect the particular language.