Audio filter bank design

1. A computer-implemented method for reducing ripple of a filter bank, the method being implemented in a computer system that includes one or more physical processors and physical storage, the method comprising:

determining a number of filters that jointly form a filter bank;

implementing, by the one or more physical processors, the determined number of filters, wherein individual filters are characterized by a center frequency and a bandwidth;

determining, by the one or more physical processors, Q-factors for the filters, wherein individual Q-factors are determined based on a ratio of the center frequency and the bandwidth;

adjusting, by the one or more physical processors, the center frequencies of the filters in the filter bank such that the filters are linearly-spaced filters;

obtaining a monotonically increasing Q-factor function for the filter bank; and

applying individual Q-factor adjustments to the individual filters based on the Q-factor function, by adjusting individual bandwidths such that corresponding individual Q-factors accord with the Q-factor function.

2. The method of claim 1 , wherein the filters in the tilter bank are infinite impulse response passband Hilbert filters.

3. The method of claim 1 , further comprising:

identifying a target frequency range for filter bank operation,

wherein adjusting the center frequencies of the filters in the filter bank is performed such that the bandwidths of the filters cover the target frequency range.

4. The method of claim 1 , wherein individual filters include an instantaneous energy output, the method further comprising:

applying an auto-regressive moving average to the instantaneous energy output of the individual filters in the filter bank.

5. The method of claim 4 , wherein the auto-regressive moving average for individual filters is characterized by attack and decay times, the method further comprising:

adjusting the auto-regressive moving average attack and decay times for the individual filters in the filter bank.

6. The method of claim 4 , further comprising:

computing, per individual filter, a maximum instantaneous energy based on the auto-regressive moving average of the instantaneous energy output;

determining a subset of filters in the filter bank having the maximum instantaneous energy above a noise threshold; and

adjusting a normalized output for the subset of filters in the filter bank having the maximum instantaneous energy above the noise threshold.

7. The method of claim 1 , further comprising:

normalizing an instantaneous energy output of the individual filters in the filter bank; and

outputting the normalized instantaneous energy output of the filters as a feature vector having a dimensionality of the number of filters.

8. A computer system for reducing ripple of a filter bank, the system comprising:

one or more physical processors; and

physical storage,

the one or more physical processors and physical storage being configured to implement:

a filter bank comprising a number of filters that jointly form the filter bank, wherein individual filters are characterized by a center frequency and a bandwidth, wherein the center-frequencies of the filters in the filter bank are adjusted such that the filters are linearly-spaced filters;

a defining module configured to define Q-factors for the filters, wherein individual O-factors are determined based on a ratio of the center frequency and the bandwidth, and wherein the defining module is further configured to obtain a monotonically increasing Q-factor function for the filter bank; and

a Q-factor adjustment module configured to apply individual Q-factor adjustments to the individual filters based on the Q-factor function, by adjusting individual bandwidths such that corresponding individual Q-factors accord with the Q-factor function.

9. The system of claim 8 , wherein the filters in the filter bank are infinite impulse response passband Hilbert filters.

10. The system of claim 8 , wherein the filters in the filter bank are implemented such that the bandwidths of the filters cover a target frequency range.

11. The system of claim 8 , wherein individual filters include an instantaneous energy output, and wherein the one or more physical processors and physical storage are further being configured to implement:

an auto-regressive moving average module configured to apply an auto-regressive moving average to the instantaneous energy output of the individual filters in the filter bank.

12. The system of claim 11 , wherein the auto-regressive moving average for individual filters is characterized by attack and decay times, and wherein the auto-regressive moving average module is further configured to adjust the auto-regressive moving average attack and decay times for the individual filters in the filter bank.

13. The system of claim 11 , wherein the one or more physical processors and physical storage are further being configured to implement:

a computing module configured to compute, per individual filter, a maximum instantaneous energy based on the auto-regressive moving average of the instantaneous energy output;

a determining module configured to determine a subset of filters in the filter bank having the maximum instantaneous energy above a noise threshold; and

an adjusting module configured to adjust a normalized output for the subset of filters in the filter bank having the maximum instantaneous energy above the noise threshold.

14. The system of claim 8 , wherein the one or more physical processors and physical storage are further being configured to implement:

a normalizer configured to normalize an instantaneous energy output of the individual filters in the filter bank and to output the normalized instantaneous energy output of the filters as a feature vector having a dimensionality of the number of filters.