A novel algorithm for audio information retrieval based on audio fingerprint

This paper proposes a novel two-step search strategy for an audio information retrieval system: a coarse search followed by an exact search. At the core of the proposed audio information retrieval system is an audio fingerprint (AF) extraction scheme based on modified discrete cosine transform (MDCT), which enables searching a large audio database with only limited computing resources and high robustness in the presence of noise. Experimental results show that the proposed AF extraction algorithm is more robust than the Philips´ AF extraction algorithm, the bit error rate (BER) is decreased 35%; the search algorithm is precise even in a noisy environment, the recall rate and precision rate are both increased 5%, and the search time is only 70% of Philips´ matching algorithm.