Music genre classification from Turkish lyrics

Abstract

The amount of music in digital form increases due to the improvement of internet and recording technologies. With this increase, the automatic organization of musics has emerged as a problem needs to be solved. For this reason, Music Information Retrieval (MIR) is commonly studied research area in recent years. In this context, with the developed Music Information Systems solution is sought for some problems such as automatic playlist creation, hit song detection, music genre or mood classification etc. In previous works, meta-data information, melodic or textual content (lyrics) of music used for feature extraction. Also, it is seen that song lyrics not commonly used and number of work in this area is not enough for Turkish. In this paper, Turkish lyrics data set created and used for automatic music genre classification. Experimental results have been conducted on support vector machines (SVM) and the effect of feature model on results has been investigated in music genre classification which considered as a classical text classification problem. The features are extracted from three different models which are Structural and Statistical Text Features (SSTF), Bag of Words (BoW) and NGram. The results shows that lyrics can be effective for Turkish music genre classification.