Noise estimation using coherence

1. A method for estimating a power spectral density of noise, the method comprising:

receiving, at one or more processing devices, an input signal representing audio captured using a microphone, the input signal comprising a first portion that represents acoustic outputs from two or more audio sources, and a second portion that represents a noise component;

iteratively modifying, by the one or more processing devices, a frequency domain representation of the input signal, such that the modified frequency domain representation represents a portion of the input signal in which effects due to the first portion are substantially reduced;

determining, from the modified frequency domain representation, an estimate of a power spectral density of the noise; and

generating a control signal configured to adjust one or more gains of an acoustic transducer corresponding one or more frequency ranges, the control signal being generated based on the estimate of the power spectral density of the noise.

2. The method of claim 1 , further comprising generating, by the one or more processing devices based on a time segment of the input signal, the frequency domain representation.

3. The method of claim 1 , wherein the frequency domain representation includes, for each frequency bin:

(i) values that each represent a level of coherence between acoustic outputs from a pair of the two or more audio sources,

(ii) values that each represent a level of coherence between an acoustic output of a particular audio source of the two or more audio sources and the audio captured using the microphone, and

(iii) values that each represent the power of the acoustic output for the particular frequency bin, of an individual audio source of the two or more audio sources.

4. The method of claim 1 , wherein the one or more gains of the acoustic transducer are adjusted to increase with an increase in the estimate of the power spectral density of the noise, and decrease with a decrease in the estimate of the power spectral density.

5. The method of claim 3 , wherein,

(i) the values that each represent a level of coherence between acoustic outputs from a pair of the two or more audio sources include one value for every permutation of pairs of the two or more audio sources,

(ii) the values that each represent a level of coherence between an acoustic output of a particular audio source of the two or more audio sources and the audio captured using the microphone include two values for each of the two or more audio sources, and

(iii) the values that each represent the power of the acoustic output for the particular frequency bin, of an individual audio source of the two or more audio sources include one value for each of the two or more audio sources.

6. The method of claim 1 , wherein the frequency domain representation comprises a cross-spectral density matrix computed based on outputs of the two or more audio sources.

7. The method of claim 6 , wherein iteratively modifying the frequency domain representation comprises executing a matrix diagonalization process on the cross-spectral density matrix.

8. The method of claim 1 , wherein the first portion represents acoustic outputs from three or more audio sources.

9. A system comprising:

a noise analysis engine comprising one or more processing devices, the noise analysis engine configured to:

receive an input signal representing audio captured using a microphone, the input signal comprising a first portion that represents acoustic outputs from two or more audio sources, and a second portion that represents a noise component;

iteratively modify a frequency domain representation of the input signal, such that the modified frequency domain representation represents a portion of the input signal in which effects due to the first portion are substantially reduced;

determine, from the modified frequency domain representation, an estimate of a power spectral density of the noise; and

generate a control signal configured to adjust one or more gains of an acoustic transducer corresponding one or more frequency ranges, the control signal being generated based on the estimate of the power spectral density of the noise.

10. The system of claim 9 , wherein the noise analysis is further configured to generate, based on a time segment of the input signal, the frequency domain representation.

11. The system of claim 9 , wherein the frequency domain representation includes, for each frequency bin:

(i) values that each represent a level of coherence between acoustic outputs from a pair of the two or more audio sources,

(ii) values that each represent a level of coherence between an acoustic output of a particular audio source of the two or more audio sources and the audio captured using the microphone, and

(iii) values that each represent the power of the acoustic output for the particular frequency bin, of an individual audio source of the two or more audio sources.

12. The system of claim 11 , wherein,

(i) the values that each represent a level of coherence between acoustic outputs from a pair of the two or more audio sources include one value for every permutation of pairs of the two or more audio sources,

(ii) the values that each represent a level of coherence between an acoustic output of a particular audio source of the two or more audio sources and the audio captured using the microphone include two values for each of the two or more audio sources, and

(iii) the values that each represent the power of the acoustic output for the particular frequency bin, of an individual audio source of the two or more audio sources include one value for each of the two or more audio sources.

13. The system of claim 9 , wherein the one or more gains of the acoustic transducer are adjusted to increase with an increase in the estimate of the power spectral density of the noise, and decrease with a decrease in the estimate of the power spectral density.

14. The system of claim 9 , wherein the frequency domain representation comprises a cross-spectral density matrix computed based on outputs of the two or more audio sources.

15. The system of claim 14 , wherein iteratively modifying the frequency domain representation comprises executing a matrix diagonalization process on the cross-spectral density matrix.

16. The system of claim 9 , wherein the first portion represents acoustic outputs from three or more audio sources.

17. One or more machine-readable storage devices having encoded thereon computer readable instructions for causing one or more processing devices to perform operations comprising:

receiving an input signal representing audio captured using a microphone, the input signal comprising a first portion that represents acoustic outputs from two or more audio sources, and a second portion that represents a noise component;

iteratively modifying a frequency domain representation of the input signal, such that the modified frequency domain representation represents a portion of the input signal in which effects due to the first portion are substantially reduced;

determining, from the modified frequency domain representation, an estimate of a power spectral density of the noise; and

generating a control signal configured to adjust one or more gains of an acoustic transducer corresponding one or more frequency ranges, the control signal being generated based on the estimate of the power spectral density of the noise.

18. The one or more machine-readable storage devices of claim 17 , further comprising generating, by the one or more processing devices based on a time segment of the input signal, the frequency domain representation.

19. The one or more machine-readable storage devices of claim 17 , wherein the frequency domain representation includes, for each frequency bin:

(i) values that each represent a level of coherence between acoustic outputs from a pair of the two or more audio sources,

(ii) values that each represent a level of coherence between an acoustic output of a particular audio source of the two or more audio sources and the audio captured using the microphone, and

(iii) values that each represent the power of the acoustic output for the particular frequency bin, of an individual audio source of the two or more audio sources.

20. The one or more machine-readable storage devices of claim 19 , wherein,

(i) the values that each represent a level of coherence between acoustic outputs from a pair of the two or more audio sources include one value for every permutation of pairs of the two or more audio sources,

(ii) the values that each represent a level of coherence between an acoustic output of a particular audio source of the two or more audio sources and the audio captured using the microphone include two values for each of the two or more audio sources, and

(iii) the values that each represent the power of the acoustic output for the particular frequency bin, of an individual audio source of the two or more audio sources include one value for each of the two or more audio sources.

21. The one or more machine-readable storage devices of claim 17 , wherein the one or more gains of the acoustic transducer are adjusted to increase with an increase in the estimate of the power spectral density of the noise, and decrease with a decrease in the estimate of the power spectral density.

22. The one or more machine-readable storage devices of claim 17 , wherein the frequency domain representation comprises a cross-spectral density matrix computed based on outputs of the two or more audio sources.

23. The one or more machine-readable storage devices of claim 22 , wherein iteratively modifying the frequency domain representation comprises executing a matrix diagonalization process on the cross-spectral density matrix.

24. The one or more machine-readable storage devices of claim 17 , wherein the first portion represents acoustic outputs from three or more audio sources.