Songs Recognition Using Audio Information Fusion

Abstract

Abstract The article presents information fusion approach for song classification with use of acoustic signal. Many acoustic features can contribute to correct identification of a song. Taking into consideration only one set of features may result in omission of relevant information. It is possible to improve the accuracy of identification process by means of the information fusion technique, in which various aspects of acoustic fingerprint are taken into consideration. Two sets of signal features were distinguished: one were based on frequency analysis (harmonic elements) and the other were based on multidimensional correlation ratios. An identification of a commercial was made with use of SVM and k-NN classifiers. The music audio signal database was used for assessing the effectiveness of the proposed solution. Results show an improved effectiveness of identification in relation to applying only one set of song features