System and method of using neural transforms of robust audio features for speech processing

1. The method comprising:

receiving, via a communication network, a first information stream associated with speech, the first information stream comprising micro-modulation features modeled in a first time scale;

receiving, via the communication network, a second information stream associated with the speech, the second information stream comprising cepstral features modeled in a second time scale, wherein the first time scale is distinct from the second time scale;

combining, via a non-linear multilayer perceptron, the first information stream and the second information stream, to yield a third information stream; and

performing, via a hardware processor, automatic speech recognition on the third confirmation stream.

2. The method of claim 1 , further comprising filtering out noise from the third information stream prior to performing automatic speech recognition.

3. The method of claim 1 , wherein the third information stream comprises less features than raw features in the first information stream and the second information stream.

4. The method of claim 1 , further comprising training a Hidden Markov model using the third information stream.

5. The method of claim 1 , wherein the combining comprises deep neural network architecture.

6. The method of claim 1 , wherein the micro-modulation features comprise a mean formant frequency.

7. The method of claim 1 , wherein the micro-modulation features comprise estimates of normalized formant frequency variances.

8. The method of claim 1 , wherein the combining of the first information stream and the second information stream, and the performing of the automatic speech recognition, occur on a non-linear discriminative front end.

9. The method of claim 1 , wherein the non-linear multilayer perceptron generates bottleneck features from acoustic features identified in the speech.

10. The method of claim 9 , wherein the bottleneck features are weighted using an iterative stochastic gradient descent and a training corpus.

11. A system comprising:

a processor; and

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

receiving, via a communication network, a first information stream associated with speech, the first information stream comprising micro-modulation features modeled in a first time scale;

receiving, via the communication network, a second information stream associated with the speech, the second information stream comprising cepstral features modeled in a second time scale, wherein the first time scale is distinct from than the second time scale;

combining, via a non-linear multilayer perceptron, the first information stream and the second information stream, to yield a third information stream; and

performing automatic speech recognition on the third confirmation stream.

12. The system of claim 11 , wherein the computer-readable medium stores further instructions which, when executed by the processor, cause the processor to perform a further operation comprising filtering out noise from the third information stream prior to performing automatic speech recognition.

13. The system of claim 11 , wherein the third information stream comprises less features than raw features in the first information stream and the second information stream.

14. The system of claim 11 , wherein the computer-readable storage medium stores further instructions which, when executed by the processor, cause the processor to perform a further operation comprising:

training a Hidden Markov model using the third information stream.

15. The system of claim 11 , wherein the combining comprises deep neural networks architecture.

16. A non-transitory computer-readable storage device storing instructions, which, when executed by a processor, cause the processor to perform operations comprising:

receiving, via a communication network, a first information stream associated with speech, the first information stream comprising micro-modulation features modeled in a first time scale;

receiving, via the communication network, a second information stream associated with the speech, the second information stream comprising cepstral features modeled in a second time scale, wherein the first time scale is distinct from than the second time scale;

combining, via a non-linear multilayer perceptron, the first information stream and the second information stream, to yield a third information stream; and

performing automatic speech recognition on the third confirmation stream.

17. The non-transitory computer-readable storage device of claim 16 , wherein the computer-readable storage device further stores instructions which, when executed by the processor, cause the processor to perform a further operation comprising filtering out noise from the third information stream prior to performing automatic speech recognition.

18. The non-transitory computer-readable storage device of claim 16 , wherein the third information stream comprises less features than raw features in the first information stream and the second information stream.

19. The non-transitory computer-readable storage device of claim 16 , wherein the computer-readable storage device stores further instructions which, when executed by the processor, cause the processor to perform a further operations comprising training a Hidden Markov model using the third information stream.

20. The non-transitory computer-readable storage device of claim 16 , wherein the combining comprises deep neural networks architecture.