Finding Audio Fingerprinter Using GPU

Abstract

To deal with multimedia objects, specially audio signals, we need to get an object representation that is stable and persistent to different natural degradation of the objects. This representation is called fingerprint signal, particulary we focused on Audio Fingerprint (AFP).
An AFP should be an invariant of the audio signal, an intrinsic characteristic found in it even if it has suffered severe degradations as long as it is still recognizable. If AFP represents the perceptual audio features, it can be used to measure the similarity between audio signals. In order to design an AFP, a dense representation is more robust than a sparse one. A dense representation also imply more compute cycles and hence a slower processing speed.
The computational power associated with dedicated technologies for specific purposes, constant development and low cost, have provided a valid alternative to parallel supercomputers. One the most popular dedicated technologies are the GPU (Graphics Processing Unit). To speedup the computing of a very dense audio fingerprint on GPU is our challenge.
In this work, we propose to obtain an audio fingerprint through the application of high perfomance computing techniques using a parallel architecture as GPU. This parallel system will be part of a comprehensive system, which will allow to determinate the audio fingerprints of several audio signals generated simultaneously. Finally, some experimental results are showed.