Fast parallel algorithm for audio content retrieval on GPUs

The search techniques audio content MIR (music information retrieval) face two major challenges: the robustness of the algorithm and the speed of this operation. In this article proposes a model of fast algorithm for the extraction of audio data by the fingerprinting technique, which is implemented on a CPU-based platform and then parallelized to run on a graphics hardware (GPU). Tests on GPU determined a success rate close to 100% and response times approximately 2 times lower than those obtained with a PC workstation, allowing searches of up to 65 commercial in real-time.