In this day and age of music streaming apps like Spotify, SoundCloud and iTunes, organization and management of music is vital. Gone are the days of CDs and other physical forms of music. With millions of songs available on the internet to download and stream, classification of music for easy retrieval is needed. One way to classify music is according to its genre, which is a conventional category that identifies some pieces of music as belonging to a shared tradition or set of conventions. However, due to the artistic nature of music genres, these classifications can be very subjective. These reasons make the task of music genre classification quite challenging in the field of Music Information Retrieval (MIR). Automatic music genre classification can be beneficial especially for online music streaming websites like Spotify and increase user engagement. In this project I aim to classify music into genres using two different machine learning methods.
- In the first part of the project, we extract time and frequency domain features from the music samples and pass those features to traditional machine learning classifiers.
- The second method aims to classify the samples by generating MEL spectrograms of the samples and performing image classification using deep learning techniques like CNNs and CRNNs. Using CRNNs, we were able to achieve an accuracy score of 86% to classify 1000 music samples into 10 genres.
For a complete overview of the data, process, implementation and its results, please read the project report from the music-genre-classification-2.pdf file uploaded in the repository.