System and method for enhancing a monaural audio signal

1. A method for enhancing a signal, the method comprising:

receiving the signal, the receiving the signal comprising converting the signal to a frequency-domain such that the signal is separated into sub-bands;

determining energy estimates of the signal for each sub-band of the signal;

calculating an energy level distribution based on the energy estimates over time, the energy level distribution being a probability distribution as a function of energy level;

determining a noise energy estimate for each sub-band of the signal and a speech energy estimate for each sub-band of the signal based on the energy level distribution;

generating a mask for each sub-band of the signal based, in part, on the noise energy estimate for each sub-band of the signal and speech energy estimate for each sub-band of the signal;

applying the mask to the signal for each sub-band of the signal to produce an enhanced signal; and

outputting the enhanced signal.

2. The method of claim 1 , wherein outputting the enhanced signal comprises converting the enhanced signal to a time-domain.

3. The method of claim 1 , wherein a distribution of speech energy in the signal is different from a distribution of noise energy in the signal.

4. The method of claim 1 , wherein speech energy in the signal varies more rapidly than noise energy in the signal.

5. A system for enhancing a signal, the system comprising:

an input device configured to receive the signal;

a frequency analysis module configured to convert the signal to a frequency-domain such that the signal is separated into sub-bands;

a features module configured to determine energy estimates of the signal for each sub-band of the signal;

an inference engine configured to determine an energy level distribution based on the energy estimates over time, the energy level distribution being a probability distribution as a function of energy level, and the inference engine being further configured to determine a noise energy estimate for each sub-band of the signal and a speech energy estimate for each sub-band of the signal based on the energy level distribution;

a mask generator engine configured to generate a mask for each sub-band of the signal based, in part, on the noise energy estimate for each sub-band of the signal and speech energy estimate for each sub-band of the signal;

a masking module configured to apply the mask for each sub-band of the signal to the signal to produce an enhanced signal; and

an output device configured to output the enhanced signal.

6. The system of claim 5 , further comprising a frequency synthesis module configured to convert the enhanced signal to a time-domain.

7. The system of claim 5 , wherein the inference engine is further configured to assume that a distribution of speech energy in the signal is different from a distribution of noise energy in the signal.

8. The system of claim 5 , wherein the inference engine is further configured to assume that speech energy in the signal varies more rapidly than noise energy in the signal.

9. A non-transitory computer readable storage medium having embodied thereon a program, the program being executable by a processor for performing a method for enhancing a signal, the method comprising:

receiving the signal, the receiving the signal comprising converting the signal to a frequency-domain such that the signal is separated into sub-bands;

determining energy estimates of the signal for each sub-band of the signal;

determining an energy level distribution based on the energy estimates over time, the energy level distribution being a probability distribution as a function of energy level;

determining a noise energy estimate for each sub-band of the signal and a speech energy estimate for each sub-band of the signal based on the energy level distribution;

generating a mask for each sub-band of the signal based, in part, on the noise energy estimate for each sub-band of the signal and speech energy estimate for each sub-band of the signal;

applying the mask to the signal for each sub-band of the signal to produce an enhanced signal; and

outputting the enhanced signal.

10. The non-transitory computer readable storage medium of claim 9 , wherein outputting the enhanced signal comprises converting the enhanced signal to a time-domain.

11. The non-transitory computer readable storage medium of claim 9 , wherein speech energy in the signal varies more rapidly than noise energy in the signal.

12. The method of claim 1 , wherein applying the mask results in attenuating signal components with low energy, and preserving signal components with high energy.

13. The method of claim 1 , wherein determining the energy level distribution comprises tracking one or more percentile estimates of the energy estimates over time.

14. The method of claim 1 , wherein calculating the energy level distribution comprises tracking an energy histogram over time.

15. The method of claim 1 , wherein determining the noise energy estimate for each sub-band of the signal and the speech energy estimate for each sub-band of the signal comprises determining a speech energy histogram and a noise energy histogram.

16. The method of claim 1 , wherein determining the noise energy estimate for each sub-band of the signal and the speech energy estimate for each sub-band of the signal depends on a predetermined speech energy histogram.

17. The method of claim 1 , wherein determining the noise energy estimate for each sub-band of the signal and the speech energy estimate for each sub-band of the signal is based on one or more differences between tracked percentile estimates.

18. The method of claim 1 , wherein determining the noise energy estimate for each sub-band of the signal is performed by computing a non-linear function of at least one of the energy estimates for the particular sub-band, a temporally-smoothed energy estimate for the particular sub-band, and a frequency-smoothed energy estimate for the particular sub-band.