METHOD AND SYSTEM FOR ANALYZING DIGITAL AUDIO FILES

• US 20110035035A1
• Filed: 10/18/2010
• Published: 02/10/2011
• Est. Priority Date: 10/24/2000
• Status: Abandoned Application

• Current Assignee: Rovi Technologies Corporation (Adeia Inc.)
• Original Assignee: Rovi Technologies Corporation (Adeia Inc.)
• Inventors: Khan, Rehan M., Tzanetakis, George
• Application Number:
  US12/906,584
• Publication Number:
  US 20110035035A1
• Time in Patent Office: Days
• US Class Current: 700/94
• CPC Class Codes: G06F 16/634 Query by example, e.g. query by humming
• CPC Class Codes: G06F 16/635 Filtering based on additional data, e.g. user or group profiles
• CPC Class Codes: G06F 16/639 using playlists
• CPC Class Codes: G06F 16/68 Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
• CPC Class Codes: G06F 16/683 using metadata automatically derived from the content
• CPC Class Codes: Y10S 707/99945 Object-oriented database structure processing
• CPC Class Codes: Y10S 707/99948 Application of database or data structure, e.g. distributed, multimedia, or image

Abstract

A fingerprint is generated from an unknown audio signal by dividing the unknown audio signal into bins, where each bin includes points representing a feature space. Each of the points is mapped to one of a plurality of predetermined cluster centers based on the distance between each point and the plurality of cluster centers, each cluster center being associated with an element of a codebook. A string of elements is generated based on the mapping and compressed.

Claims

1. A method of generating a fingerprint from an unknown audio signal, the method comprising:

dividing the unknown audio signal into bins, each bin including a plurality of points representing a feature space;

mapping each of the plurality of points to one of a plurality of predetermined cluster centers based on the distance between each point and the plurality of cluster centers, each cluster center being associated with an element of a codebook;

generating a string of elements based on the mapping; and

compressing the string of elements.

2. The method according to claim 1, further comprising:

performing a Fourier transform on the plurality of bins to generate a frequency representation of each bin; and

computing at least one feature from the frequency representation of each bin.

3. The method according to claim 2, wherein the at least one feature includes at least one of:

(a) a spectral centroid;

(b) a spectral rolloff;

(c) a spectral flux;

(d) a peak ratio;

(e) a subband energy vector;

(f) a subband flux; and

(g) a subband energy ratio.

4. The method according to claim 1, wherein each point representing the feature space specifies a location in a multidimensional feature space.

5. The method according to claim 1, wherein the plurality of cluster centers are determined by using a c-means clustering algorithm.

6. The method according to claim 1, wherein the distance between each point and the plurality of cluster centers is based on a Mahalonobis distance.

7. The method according to claim 1, wherein each element of the codebook corresponds to a unique ASCII character.

8. The method according to claim 1, further comprising:

parsing the compressed set of elements into a plurality of sub-strings of a predefined length of characters.

matching each substring to a database of precomputed signatures to obtain at least one base signature;

assigning a match index value (MI) to each match obtained by the matching given by;

MI=Σ

W_i*N_i,

where W_i is a weight for matches of length i and N_i is a number of substring matches of length i; and

selecting a best match based on the match index value (MI).

9. A non-transitory computer-readable medium storing a computer instructions, which when executed by one or more processors, causes the one or more processors to execute the steps of:

dividing the unknown audio signal into bins, each bin including a plurality of points representing a feature space;

mapping each of the plurality of points to one of a plurality of predetermined cluster centers based on the distance between each point and the plurality of cluster centers, each cluster center being associated with an element of a codebook;

generating a string of elements based on the mapping; and

compressing the string of elements.

10. The non-transitory computer-readable medium according to claim 9, further comprising the steps of:

performing a Fourier transform on the plurality of bins to generate a frequency representation of each bin; and

computing at least one feature from the frequency representation of each bin.

11. The non-transitory computer-readable medium according to claim 10, wherein the at least one feature includes at least one of:

(a) a spectral centroid;

(b) a spectral rolloff;

(c) a spectral flux;

(d) a peak ratio;

(e) a subband energy vector;

(f) a subband flux; and

(g) a subband energy ratio.

12. The non-transitory computer-readable medium according to claim 9, wherein each point representing the feature space specifies a location in a multidimensional feature space.

13. The non-transitory computer-readable medium according to according to claim 9, wherein the plurality of cluster centers are determined by using a c-means clustering algorithm.

14. The non-transitory computer-readable medium according to according to claim 9, wherein the distance between each point and the plurality of cluster centers is based on a Mahalonobis distance.

15. The non-transitory computer-readable medium according to according to claim 9, wherein each element of the codebook corresponds to a unique ASCII character.

16. The non-transitory computer-readable medium according to according to claim 9, further comprising the steps of:

parsing the compressed set of elements into a plurality of sub-strings of a predefined length of characters.

matching each substring to a database of precomputed signatures to obtain at least one base signature;

assigning a match index value (MI) to each match obtained by the matching given by;

MI=Σ

W_i*N_i,

where W_i is a weight for matches of length i and N_i is a number of substring matches of length i; and

selecting a best match based on the match index value (MI).

17. An apparatus for generating a fingerprint from an unknown audio signal, comprising:

at least one processor operable to perform;

dividing the unknown audio signal into bins, each bin including a plurality of points representing a feature space;

mapping each of the plurality of points to one of a plurality of predetermined cluster centers based on the distance between each point and the plurality of cluster centers, each cluster center being associated with an element of a codebook stored in a memory;

generating a string of elements based on the mapping; and

compressing the string of elements.

18. The apparatus according to claim 17, the processor further operable to perform a Fourier transform on the plurality of bins to generate a frequency representation of each bin and compute at least one feature from the frequency representation of each bin.

19. The apparatus according to claim 18, wherein the at least one feature includes at least one of:

(a) a spectral centroid;

(b) a spectral rolloff;

(c) a spectral flux;

(d) a peak ratio;

(e) a subband energy vector;

(f) a subband flux; and

(g) a subband energy ratio.

20. The apparatus according to claim 17, wherein each point representing the feature space specifies a location in a multidimensional feature space.

21. The apparatus according to claim 17, wherein the plurality of cluster centers are determined by using a c-means clustering algorithm.

22. The apparatus according to claim 17, wherein the distance between each point and the plurality of cluster centers is based on a Mahalonobis distance.

23. The method according to claim 17, wherein each element of the codebook corresponds to a unique ASCII character.

24. The apparatus according to claim 17, wherein the at one processor is further operable to perform the following steps:

parsing the compressed set of elements into a plurality of sub-strings of a predefined length of characters.

matching each substring to a database of precomputed signatures to obtain at least one base signature;

assigning a match index value (MI) to each match obtained by the matching given by;

MI=Σ

W_i*N_i,

where W_i is a weight for matches of length i and N_i is a number of substring matches of length i; and

selecting a best match based on the match index value (MI).