Media content identification on mobile devices

1. A method for audio fingerprinting by using content-based audio feature extraction, the method comprising:

analyzing input audio samples, divided into overlapping frames, to produce windowed audio frame samples for each overlapped frame;

computing a fast Fourier transform (FFT) for the windowed audio frame samples which FFT results are filtered by a filter-bank on the spectral components to produce a set of filter bank output coefficients;

computing a log function and a square root function of each set of filter bank output coefficients to produce log and square root output coefficients;

applying a discrete cosine transform (DCT) separately to the log and square root output coefficients to produce two sets of DCT spectral components; and

applying a temporal multi-tap finite impulse response (FIR) smoothing derivative filter to the two sets of DCT spectral components to produce two separate primary descriptors, wherein values in the two separate primary descriptors are sampled to produce two primary signatures.

2. The method of claim 1 further comprising:

applying a threshold to each dimension of the two separate primary descriptors to derive two separate primary signatures.

3. The method of claim 1 , wherein the logarithm of the output coefficients from each filter bank is computed to handle a wider range of volume levels.

4. The method of claim 1 , wherein the square root of the output coefficients from each filter bank is computed to handle higher levels of noise.

5. The method of claim 1 , wherein the DCT is applied to the log output coefficients to convert the log output coefficients to a new set of magnitude levels and frequencies.

6. The method of claim 1 , wherein the DCT is applied to the square root output coefficients to convert the square root output coefficients to a new set of magnitude levels and frequencies.

7. The method of claim 1 , wherein the temporal multi-tap FIR smoothing derivative filter is applied on each of the DCT spectral components in a temporal domain of the audio signature generation process to produce the two separate primary descriptors.

8. The method of claim 1 , wherein a number of lower coefficients of primary descriptors are extracted, and then combined with additional bits from a set of separately detected audio content features to form data base cluster signatures.

9. The method of claim 1 , wherein a number of lower coefficients of primary descriptors are extracted, and then combined with additional bits from a set of separately detected video content features to form data base cluster signatures.

10. The method of claim 1 further comprising:

adjusting filters of the filter-bank according to center frequency measures of the audio frame samples prior to filtering the FFT results.

11. The method of claim 1 further comprising:

determining a center frequency of a current audio frame of the audio input;

determining which filter in the filter-bank has a center frequency closest to the determined center frequency of the current audio frame as compared to other filters in the filter-bank;

generating a multiplier factor to scale the determined filter's center frequency to the determined center frequency of the current audio frame; and

adjusting the filters in the filter-bank by the generated multiplier factor to create an adjusted filter-bank to filter the FFT results.

12. The method of claim 1 further comprising:

applying the filter-bank on K frequency bands to produce K output coefficients that are divided into m linearly spaced bands and n logarithmically spaced bands.

13. The method of claim 1 further comprising:

applying the filter-bank on K frequency bands to produce K output coefficients that are provided as input to a square root function to produce final filter-bank outputs.

14. The method of claim 1 , wherein the temporal multi-tap FIR smoothing derivative filter is a 9 tap FIR smoothing derivative filter.

15. The method of claim 14 , wherein the 9 tap FIR smoothing derivative filter generates 9 FIR filter coefficients which are in a (−1, 1) interval.

16. The method of claim 1 further comprising:

quantizing each coefficient in each set of the two sets of DCT spectral components to either a zero or a one based on a sign of the coefficient; and

concatenating the quantized coefficients from each set of the DCT spectral components to produce type A signatures that correspond to the log function on the filter-bank outputs and type B signatures that correspond to the square root function on the filter-bank outputs.

17. The method of claim 16 further comprising:

selecting the type A signatures and the type B signatures that have at least k zeros and k ones, wherein k is a predetermined value.

18. The method of claim 16 further comprising:

suppressing the type A signatures and the type B signatures that have less than k zeros and k ones, wherein k is a predetermined value.

19. The method of claim 16 further comprising;

combining the type A signatures with the type B signatures for storage in a signature database on a portable media device.

20. The method of claim 1 , wherein the input audio samples include speech and music interference received from multiple sources external to a portable media device that processes the input audio samples.