Method and system for robust language identification

1. A method for determining a language identity corresponding to a speech signal, comprising:

receiving the speech signal;

partitioning the speech signal into a plurality of audio frames;

extracting features of the plurality of audio frames;

determining, using a neural network, a variable associated with the language identity and one or more auxiliary attributes of the speech signal, for each of the plurality of audio frames;

determining scores of the plurality of audio frames based on the extracted features; and

determining the language identity of the speech signal based on the variables and scores determined for the plurality of audio frames by:

computing a weighted average of the variables based on the scores of the plurality of audio frames; and

determining the language identity of the speech signal based on the weighted average.

2. The method of claim 1 , wherein the variable indicates a probability associated with the language identity.

3. The method of claim 1 , wherein the auxiliary attributes correspond to one or more of speaker gender, distance between a sound source and a microphone, background information, noise condition, channel information, or microphone information.

4. The method of claim 1 , wherein the neural network is a convolutional neural network.

5. The method of claim 1 , wherein the extracted features include filter-bank (FBANK) features, i-Vectors, or Mel-Frequency Cepstral Coefficients (MFCCs).

6. The method of claim 1 , wherein determining the scores of the plurality of audio frames based on the extracted features includes:

extracting features from training data corresponding to a plurality of languages;

performing a regression analysis on the features extracted from the training data and language identities of the training data;

constructing a scoring function based on the regression analysis; and

determining the scores of the plurality of audio frames based on the scoring function.

7. The method of claim 1 , wherein determining the scores of the plurality of audio frames based on the extracted features includes:

extracting features from training data corresponding to a plurality of languages;

clustering the features extracted from the training data into a plurality of clusters;

performing classification of the extracted features of the plurality of audio frames into the plurality of clusters; and

determining the scores of the plurality of audio frames based on the classification.

8. A system for determining a language identity corresponding to a speech signal, the system comprising:

a memory including instructions; and

a processor configured to execute the instructions to:

receive the speech signal;

partition the speech signal into a plurality of audio frames;

extract features of the plurality of audio frames;

determine, using a neural network, a variable associated with the language identity and one or more auxiliary attributes of the speech signal, for each of the plurality of audio frames;

determine scores of the plurality of audio frames based on the extracted features; and

determine the language identity of the speech signal based on the variables and scores determined for the plurality of audio frames by:

computing a weighted average of the variables based on the scores of the plurality of audio frames; and

determining the language identity of the speech signal based on the weighted average.

9. The system of claim 8 , wherein the variable indicates a probability associated with the language identity.

10. The system of claim 8 , wherein the auxiliary attributes correspond to one or more of speaker gender, distance between a sound source and a microphone, background information, noise condition, channel information, or microphone information.

11. The system of claim 8 , wherein the neural network is a convolutional neural network.

12. The system of claim 8 , wherein the extracted features include filter-bank (FBANK) features, i-Vectors, or Mel-Frequency Cepstral Coefficients (MFCCs).

13. The system of claim 8 , wherein the processor is further configured to execute the instructions to:

extract features from training data corresponding to a plurality of languages;

perform a regression analysis on the features extracted from the training data and language identities of the training data;

construct a scoring function based on the regression analysis; and

determine the scores of the plurality of audio frames based on the scoring function.

14. The system of claim 8 , wherein the processor is further configured to execute the instructions to:

extract features from training data corresponding to a plurality of languages;

cluster the features extracted from the training data into a plurality of clusters;

perform classification of the extracted features of the plurality of audio frames into the plurality of clusters; and

determine the scores of the plurality of audio frames based on the classification.

15. A non-transitory computer-readable medium storing instructions that, when executed by one or more processors, cause the processors to perform a method for determining a language identity corresponding to a speech signal, the method comprising:

receiving the speech signal;

partitioning the speech signal into a plurality of audio frames;

extracting features of the plurality of audio frames;

determining, using a neural network, a variable associated with the language identity and one or more auxiliary attributes of the speech signal, for each of the plurality of audio frames;

determining scores of the plurality of audio frames based on the extracted features; and

determining the language identity of the speech signal based on the variables and scores determined for the plurality of audio frames by:

computing a weighted average of the variables based on the scores of the plurality of audio frames; and

determining the language identity of the speech signal based on the weighted average.

16. The non-transitory computer-readable medium of claim 15 , wherein determining the scores of the plurality of audio frames based on the extracted features includes:

extracting features from training data corresponding to a plurality of languages;

performing a regression analysis on the features extracted from the training data and language identities of the training data;

constructing a scoring function based on the regression analysis; and

determining the scores of the plurality of audio frames based on the scoring function.

17. The non-transitory computer-readable medium of claim 15 , wherein determining the scores of the plurality of audio frames based on the extracted features includes:

extracting features from training data corresponding to a plurality of languages;

clustering the features extracted from the training data into a plurality of clusters;

performing classification of the extracted features of the plurality of audio frames into the plurality of clusters; and

determining the scores of the plurality of audio frames based on the classification.