US 9,684,715 B1Granted

Audio identification using ordinal transformation

AssigneeGoogle Inc.

InventorsDavid Ross, Jay Yagnik

Life of this patent

Estimating the term…

Loading patent…

Abstract

This disclosure relates to audio identification using ordinal transformations. A media matching component receives a sample audio file. The sample audio file can include, for example, a cover song. The media matching component includes a vector component that computes a set of vectors using auditory feature values included in the sample audio file. A hashing component employs a hash function to generate a fingerprint, including a set of sub-fingerprints, for the sample audio file using the set of vectors. The fingerprint is invariant to variations including but not limited to variations in key, instrumentation, encoding formats, performers, performance conditions, arrangement, and/or recording and processing variations. An identification component determines if any reference audio files are similar to the sample audio file using the fingerprint and/or sub-fingerprints, and identifies any similar reference audio files.

Claims22

Dependent claims

A computer recognition system for performing media matching, comprising:

a memory storing computer executable components;
and
a processor configured to execute the following computer executable components stored in the memory:
a receipt component that receives a sample audio file containing unknown audio content uploaded to a media hosting service using a computer network;
a vector component that computes a set of vectors using auditory feature values identified in the sample audio file;
a hash component that employs a winner take all hash function for non-binary vectors to generate sub-fingerprints for corresponding respective vectors in the set of vectors, wherein the winner take all hash function for each vector X in the set of vectors:
generates a set of m permutations Θ, where m is an integer;
for each permutation Θi where i is an integer from 0 to m−1:
permutates elements of vector X according to Θi to generate permutated vector X′ having K non-binary elements, where K is an integer,
initializes sparse code cxi to 0, and
sets cxi to an element of maximum value in permutated vector X′;
and
sets subfingerprint Cx equal to (cx0, cx1, . . . cxm−1);
and
an identification component that:
identifies at least one reference audio file containing known audio content similar to the sample audio file using the sub-fingerprints, the at least one reference audio file stored in a database in communication with the media hosting service;
determines an identity of the audio content contained in the sample audio file responsive to the identified at least one reference audio file;
and
detects that the audio content contained in the sample audio file is a transformed performance of the known audio content responsive to the determined identity.

16 more elements

A method of performing media matching, comprising:

receiving, by a system including a processor, a sample audio file containing unknown audio content uploaded to a media hosting service using a computer network;
computing, by the system, a set of vectors using auditory feature values identified for the sample audio file;
employing, by the system, a winner-take-all hash function for non-binary vectors to generate sub-fingerprints corresponding to respective vectors in the set of vectors, wherein employing the winner take all hash function for each vector X in the set of vectors comprises:
generating a set of m permutations Θ, where m is an integer;
for each permutation Θi where i is an integer from 0 to m−1:
permutating elements of vector X according to Θi to generate permutated vector X′ having K non-binary elements, where K is an integer,
initializing sparse code cxi to 0, and
setting cxi to an element of maximum value in permutated vector X′;
and
setting subfingerprint Cx equal to (cx0, cx1, . . . cxm−1);
and
identifying, by the system, at least one reference audio file containing known audio content similar to the sample audio file based on a set of similarity criteria using the sub-fingerprints, the at least one reference audio file stored in a database in communication with the media hosting service;
determining an identity of the audio content contained in the sample audio file responsive to the identified at least one reference audio file;
and
detecting that the audio content contained in the sample audio file is a transformed performance of the known audio content responsive to the determined identity.

12 more elements

A non-transitory computer-readable medium having instructions stored thereon computer executable instructions that, in response to execution, cause a system including a processor to perform operations comprising:

receiving a sample audio file containing unknown audio content uploaded to a media hosting service using a computer network;
computing a set of vectors using auditory feature values identified for the sample audio file;
employing a winner-take-all hash function for non-binary vectors to generate sub-fingerprints corresponding to respective vectors in the set of vectors, wherein employing the winner take all hash function for each vector X in the set of vectors comprises:
generating a set of m permutations Θ, where m is an integer;
for each permutation Θi, where i is an integer from 0 to m−1:
permutating elements of vector X according to Θi to generate permutated vector X′ having K non-binary elements, where K is an integer,
initializing sparse code cxi to 0, and
setting cxi to an element of maximum value in permutated vector X′;
and
setting subfingerprint Cx equal to (cx0, cx1, . . . cxm−1);
and
identifying at least one reference audio file containing known audio content similar to the sample audio file based on a set of similarity criteria using the sub-fingerprints, the at least one reference audio file stored in a database in communication with the media hosting service;
determining an identity of the audio content contained in the sample audio file responsive to the identified at least one reference audio file;
and
detecting that the audio content contained in the sample audio file is a transformed performance of the known audio content responsive to the determined identity.

12 more elements

Dependencies are inferred from the claim text.