Wei Liu

  Intelligent Computing and Machine Learning Lab, ASEE, Beihang University

We proposed a new model for music genre classification. The model is referred to as The Bag-of-Tones (BOT) model which follows the conceptually similar idea of the bag-of-words(BOW) model in natural language processing and the bag-of-feature(BOF) model in image processing. The basic low-level music features such as Mel-frequency cepstral coefficients (MFCC) are clustered into a set of codewords referred to as "tones". By using such a model, each piece of music can be represented by a new feature vector of distribution on tones. Classical machine learning models such as support vector machines (SVM) can be applied for genre classification. By comparing to the previous work, we found the new proposed model outperform classical models on a given benchmark dataset. In general, this model can be used to structure the large collections of music available on the web. It can play an important role in automatic music categorization and retrieval.

[Download BUAA5 Dataset] <It could be too big to be downloaded, please contact me if you hope to test on the data!>

[Download GTZAN Dataset]