Instrument independent musical genre classification using random 3000 ms segment

  • Authors:

  • Ali Cenk Gedik;Adil Alpkocak

  • Affiliations:

  • Department of Musicology, Dokuz Eylul University, İzmir, Narlıdere, Turkey;Department of Computer Engineering, Dokuz Eylul University, Buca, İzmir, Turkey

  • Venue:
  • TAINN'05 Proceedings of the 14th Turkish conference on Artificial Intelligence and Neural Networks
  • Year:
  • 2005

Quantified Score

Hi-index 0.00

Visualization

Abstract

This paper presents a new approach to musical genre classification by using randomly selected maximum 3000 ms long segment of each recording. Recordings are classified into three root genres as jazz, classical and pop. A k-Nearest Neighbor (k-NN) classifier with pitch, harmony and rhythm features, independent from instruments, and a training set of 180 MIDI recordings is used for classification. The classifier is evaluated both quantitatively and cognitively using 45 MIDI recordings from outside the training set and considered for specificity, selectivity and accuracy measures. Experiments have demonstrated the good accuracy of the classifier and its high potential of use.