Audio Fingerprinting Method for Byzantine Hymn Recognition

In this work, a method for Byzantine hymn recognition is proposed utilizing an audio fingerprinting technique to generate speed and scale invariant fingerprints. The main steps of the process are the two-dimensional Fourier Transform (2DFT) and the calculation of a distance metric based on similarity matrices. The general concept is that by creating a database of fingerprints, it is possible to identify a query recording by extracting its fingerprints and compare them to the database. The method was tested by using a dataset of 351 different recordings of nine Byzantine hymns. Some key parameters of the process were evaluated in terms of accuracy and high scores, up to 99.44%, were achieved by introducing a statistical polling approach for the correct match selection.