Audio Fingerprinting with Higher Matching Depth at Reduced Computational Complexity

In this paper, a novel audio fingerprint technique with higher matching depth at a reduced computational complexity has been proposed. Any audio clip can be identified from a huge audio collection by its audio fingerprint which contains some unique extractable and perceivable features of it. One of the major concern in the process of identification of a sample audio clip from huge audio collection is the computational complexity involved in it. In proposed technique, peak pairs are chosen from the alternate time bins sorted in the descending order of the amplitude of the spectrogram of the audio file under consideration, as opposed to the sequential selection across consecutive time bins in the decreasing order of their amplitude in the basic methods. This paper also proposes a simple mode based ranking algorithm to provide the user with multiples matches in case of a possible false positive match.