Mood-based Music Genre Classification
|Projektverantwortung vor Ort:||PD Dr. Jonghwa Kim|
In this project, we will develop a novel music taxonomy that combines the four popular music categories (classical, jazz, pop, rock) with four quadrants models of musical mood derived from two factors: valence (happy/anxious) and arousal (calm/energetic).
With the rapid growth of digital music databases, it has become increasingly important to automate the task of music genre classification. In this project, we will develop a novel music taxonomy that combines the four popular music categories (classical, jazz, pop, rock) with four quadrants models of musical mood derived from two factors: valence (happy/anxious) and arousal (calm/energetic), [Thayer, 1989]. In addition to intensity features which are more suitable for the calculation of arousal, we use further rhythmic and timbral features to estimate valence. For the rhythmic features, we derive beat, tempo, drum presence and silence ratio from each octave-scaled subband of the audio signal. To obtain the timbral features, we use the mel-frequency ceptral coefficients (MFCC), linear prediction coefficient (LPC), pitch histogram and spectral energy. To classify music clips into the proposed categories, several supervised machine learning methods (e.g. GMM, k-NN, MLP, Bayesian) will be evaluated and optimized by an adaptive feature selection method to improve the classification quality, especially for ambiguous categories, e.g. rock and pop.
- Thayer, R.E. The Biopsychology of Mood and Arousal. New York: Oxford University Press, 1989
- Pachet, F and Cazaly, D., A Taxonomy of Musical Genres. Content-Based Multimedia Information Access Conference (RIAO), Paris, April 2000
- Liu, D., Lu, L. and Zhang, H.J., Automatic mood detection from acoustic music data, International Symposium on Music Information Retrieval, Baltimore, Maryland (USA), 2003
- Feng, Y. Zhuang, Y. and Pan, Y. Music Information Retrieval by Detecting Mood via Computational Media Aesthetics, IEEE/WIC International Conference on Web Intelligence, Halifax, Canada, October, 2003
- Tzanetakis, G. and Cool, P. Musical Genre Classification of Audio Signals. IEEE Transactions on Speech and Audio Processing, 10(5): 293-302, 2002
Related Bachelor-Thesis (german):