Method and apparatus for identifying an unknown work

• US 20070074147A1
• Filed: 09/28/2005
• Published: 03/29/2007
• Est. Priority Date: 09/28/2005
• Status: Active Grant

• Current Assignee: Audible Magic Corporation
• Original Assignee: Audible Magic Corporation
• Inventors: Wold, Erling
• Granted Patent: US 7,529,659 B2
• Time in Patent Office: Days
• US Class Current: 717/100
• CPC Class Codes: G06F 16/634 Query by example, e.g. query by humming
• CPC Class Codes: G06F 16/683 using metadata automatically derived from the content

Abstract

A system for determining an identity of a received work. The system receives audio data for an unknown work. The audio data is divided into segments. The system generates a signature of the unknown work from each of the segments. Reduced dimension signatures are then generated at least a portion of the signatures. The reduced dimension signatures are then compared to reduced dimensions signatures of known works that are stored in a database. A list of candidates of known works is generated from the comparison. The signatures of the unknown works are then compared to the signatures of the known works in the list of candidates. The unknown work is then identified as the known work having signatures matching within a threshold.

Claims

1. A product for determining an identity of a received work comprising:

instructions for directing a processing unit to;

receive audio data for an unknown work,

divide said audio data into a plurality of segments,

generate a plurality of signatures of said unknown work wherein each signature is generated from one of said plurality of segments,

generate a plurality of reduced dimension signatures of said unknown works wherein each of said plurality of reduced dimension signatures is generated from one of at least a portion of said plurality of signatures,

compare said plurality of reduced dimension signatures to at least one reduced dimension signature for each one of a portion of a plurality of known works having a record stored in a works database wherein each record includes an identification of said work and at least one signature of said work,

determine a list of candidates from said plurality of known works responsive to said comparisons,

compare said plurality of signatures to at least a portion of said plurality of signatures of each of said plurality of known works in said list of candidates;

determine one of said plurality of known works in said list of candidates that matches said unknown work from said comparison, and

identify said unknown work as said one of said plurality of known works that matches said unknown work; and

a media readable by said processing unit that stores said instructions.

2. The product of claim 1, wherein said instructions to generate said plurality of signatures comprises:

instructions for directing said processing unit to;

calculate a plurality of mel frequency cepstral coefficients (MFCCs) for each of a plurality segments to generate each of said plurality of signatures.

3. The product of claim 1, wherein said instructions to generate said plurality of signatures comprises:

instructions for directing said processing unit to;

calculate each of said plurality of signatures using a plurality of acoustical features from one of said plurality of segments selected from a group consisting of loudness, pitch, brightness, bandwidth, spectrum and mel frequency cepstral coefficients (MFCCs).

4. The product of claim 1 further comprising:

instructions for directing said processing unit to index said records of said plurality of known references in said database.

5. The product of claim 4 wherein said instructions to index is performed by using an indexing strategy chosen from the group consisting of:

the k-d tree, the SS-tree, the R-tree, and the SR-tree.

6. The product of claim 1, wherein each of said plurality of segments has a segment size of approximately 0.5 to 3 seconds.

7. The product of claim 1, wherein each of said plurality of segments has a segment size of approximately 1 second.

8. The product of claim 1, wherein a hop size between consecutive ones of said plurality of segments is less than 50% of a segment size of each of said plurality of segments.

9. The method of claim 1, wherein a hop size between consecutive ones of said plurality of segments is approximately 0.1 seconds.

10. The product of claim 1, wherein said instructions to generate said plurality of signatures comprises:

instructions for directing said processing unit to;

calculate an average for each of a plurality of acoustical features selected from a group consisting of;

loudness, pitch, brightness, bandwidth, spectral features, and mod frequency cepstral coefficients.

11. The product of claim 1, wherein said instructions to generate said plurality of reduced dimension signatures comprises:

instructions for directing said processing unit to;

project features of each said portion of said plurality of signatures in a Karhunen-Loeve basis.

12. The product of claim 11 wherein said instructions to generate said plurality of reduced dimension signatures comprises:

instructions for directing said processing unit to;

select a predetermined one of a plurality of mid frequency cepstral coefficients from each one of said portion of said plurality of signatures.

13. The product of claim 11, wherein said list of candidates includes a predetermined number of said plurality of known works.

14. A computer readable medium carrying one or more instructions for determining an identity of a received work, the one more instructions including instructions which executed by one or more processors, cause the one or more processors to perform the method comprising:

receiving audio data for an unknown work;

dividing said audio data into a plurality of segments;

generating a plurality of signatures of said unknown work wherein each signature is generated from one of said plurality of segments;

generating a plurality of reduced dimension signatures of said unknown works wherein each of said plurality of reduced dimension signatures is generated from one of at least a portion of said plurality of signatures;

comparing said plurality of reduced dimension signatures to at least one reduced dimension signature for each one of a portion of a plurality of known works having a record stored in a works database wherein each record includes a identification of said work and at least one signature of said work;

determining a list of candidates from said plurality of known works responsive to said comparisons;

comparing said plurality of signatures to at least a portion of said plurality of signatures of each of said plurality of known works in said list of candidates; and

determining one of said plurality of known works in said list of candidates that matches said unknown work from said comparison.

15. The method of claim 14, wherein said step for generating said plurality of signatures comprises:

calculating a plurality of mel frequency cepstral coefficients (MFCCs) for each of a plurality segments to generate each of said plurality of signatures.

16. The method of claim 14, wherein said step for generating said plurality of signatures comprises:

calculating each of said plurality of signatures using a plurality of acoustical features from one of said plurality of segments selected from a group consisting of loudness, pitch, brightness, bandwidth, spectrum and mel frequency cepstral coefficients (MFCCs).

17. The method of claim 14 further comprising:

indexing said records of said plurality of known references in said database.

18. The method of claim 17 wherein said indexing is performed by using an indexing strategy chosen from the group consisting of:

the k-d tree, the SS-tree, the R-tree, and the SR-tree.

19. The method of claim 14, wherein each of said plurality of segments has a segment size of approximately 0.5 to 3 seconds.

20. The method of claim 14, wherein each of said plurality of segments has a segment size of approximately 1 second.

21. The method of claim 14, wherein a hop size between consecutive ones of said plurality of segments is less than 50% of a segment size of each of said plurality of segments.

22. The method of claim 14, wherein a hop size between consecutive ones of said plurality of segments is approximately 0.1 seconds.

23. The method of claim 14, wherein said step for generating said plurality of signatures comprises:

calculating an average for each of a plurality of acoustical features selected from a group consisting of;

loudness, pitch, brightness, bandwidth, spectral features, and mod frequency cepstral coefficients.

24. The method of claim 14, wherein said step for generating said plurality of reduced dimension signatures comprises:

projecting features of each said portion of said plurality of signatures in a Karhunen-Loeve basis.

25. The method of claim 14 wherein said step for generating said plurality of reduced dimension signatures comprises:

selecting predetermined ones of a plurality of mid frequency cepstral coefficients from each one of said portion of said plurality of signatures.

26. The method of claim 14, wherein said list of candidates includes a predetermined number of said plurality of known works.