Ensemble interest point detection for audio matching

摘要

Systems and methods for audio matching are disclosed herein. In one embodiment, a system includes both interest point mixing and fingerprint mixing by using multiple interest point detection methods in parallel. Since multiple interest point detection methods are used in parallel, accuracy of audio matching is improved across a wide variety of audio signals. In addition the scalability of the disclosed audio matching system is increased by matching the fingerprint of an audio sample with a fingerprint of a reference sample versus matching an entire spectrogram. Accordingly, a more accurate and more general solution to audio matching can be accomplished.

主权项

1. A system, comprising:
a memory that stores computer executable components; and a processor that executes the following computer executable components stored within the memory;
an interest point detection component that concurrently employs a first interest point detection method to generate a first set of interest points for an audio sample and a second interest point detection method to generate a second set of interest points for the audio sample;a classification component that probes the audio sample to determine a classification of the audio sample that includes at least a type of media capable device that provided the audio sample;a mixing component that generates a mixed set of interest points for the audio sample that comprises a subset of the first set of interest points and a subset of the second set of interest points, wherein the subset of the first set of interest points is determined based on the classification of the audio sample and information associated with the first interest point detection method, and the subset of the second set of interest points is determined based on the classification of the audio sample and other information associated with the second interest point detection method, wherein the mixing component generates the mixed set of interest points by assigning a first weight to the first set of interest points based on the classification of the audio sample and the information associated with the first interest point detection method, and assigning a second weight to the second set of interest points based on the classification of the audio sample and the other information associated with the second interest point detection method;a fingerprint component that generates a fingerprint of the audio sample based on the mixed set of interest points; anda matching component that identifies the audio sample based on a comparison between the fingerprint of the audio sample and one or more reference fingerprints.