Week 2— Music Genre Classification with Machine Learning

Hi everyone! Welcome to second blog of our project. Our project members are Osman Fatih Çakır, Ebru Yardım and Batuhan Şenyüzlü. We are students at Hacettepe University.

In this blog I am going to talk about the progress of our research. In this week we searched articles, blogs, experiments, etc. about our project. There are other projects or researches which use GTZAN dataset and a deep learning algorithm for classify music genre.

Firstly, we want to use GTZAN genre classification dataset for our project dataset. This dataset has 10 music genres with 1000 audio files having 30 seconds duration. Each genre has 100 audio files.

Since this dataset is a trending and famous dataset, it would be good to check projects and analysis that other people have done using this dataset.

By Tao Feng from University of Illinois has an experiment and article which you can see from here. In this article GTZAN dataset is used for training. The result of classification rates are over %97.

Another article from Miguel Flores Ruiz de Eguino, Stanford University. You can read the article from here. This article uses two datasets: GTZAN and MagnaTagATune. The purpose to choose GTZAN is that GTZAN dataset has many papers. MagnaTagATune dataset is bigger than GTZAN but also has layers and other features like whether it has drums, guitar, voice, is a happy song, etc. In this research, GTZAN dataset has more accuracy than MagnaTagATune dataset.

After this research, it is more comfortable to continue to our project.

Thank you for read, see you next week!