Fast Audio Fingerprint Search Strategy for Song Identification

In this paper we present an audio fingerprinting (AF) system for song identification. For the high dimensional audio fingerprint data, two AF searching algorithms were proposed and implemented: Principle Component Analysis (PCA) and the summation of the corresponding data between different frames. The experimental results show that applying PCA algorithm, the accuracy is 94.98% while the search time is as low as 8.42%; applying Sum algorithm, the accuracy is 95.92% while the search time is as low as 3.72%. A top-N Approximate Nearest Neighbor (ANN) searching is applied to the dimension-reduced data before the exact search of N full dimension data. The final realization shows an accuracy rate of 96.11% with 8.85% of search time compared with the full search method.