Automatic Music Genre Classification Using Hybrid Genetic Algorithms
- Cite this paper as:
- Karkavitsas G.V., Tsihrintzis G.A. (2011) Automatic Music Genre Classification Using Hybrid Genetic Algorithms. In: Tsihrintzis G.A., Virvou M., Jain L.C., Howlett R.J. (eds) Intelligent Interactive Multimedia Systems and Services. Smart Innovation, Systems and Technologies, vol 11. Springer, Berlin, Heidelberg
This paper aims at developing an Automatic Music Genre Classification system and focuses on calculating algorithms that (ideally) can predict the music class in which a music file belongs. The proposed system is based on techniques from the fields of Signal Processing, Pattern Recognition, and Information Retrieval, as well as Heuristic Optimization Methods. One thousand music files are used for training and validating the classification system. These files are distributed equally in ten classes. From each file, eighty one (81) features are extracted and used to create 81 similarity matrices. These 81 similarity matrices constitute the training instances. During the training phase, feature selection takes place via a modified hybrid Genetic Algorithm in order to improve the class discrimination clarity and reduce the calculating cost. In this algorithm, the crossover probability is replaced by a parent pair number that produces new solutions via a taboo list. Also, an adaptive mutation, an adaptive local exhaustive search and an adaptive replace strategy are used, depending on whether the system has reached a local extreme. Local exhaustive search takes place in the most optimal up to the current solution neighboring chromosomes. The Genetic Algorithm fitness function constitutes a weighted nearest neighbors classifier. Thus, the chromosome fitness is proportional to the classifier accuracy that the chromosome creates. During the classification phase, the features selected via the Genetic Algorithm create an adjusted nearest neighbor classifier that performs the classifications. From each new music file pending classification, selected features are extracted and then compared with the corresponding features of the database music files. The music file is assigned to the class indicated by the k nearest music files.
KeywordsSignal Processing Pattern Recognition Music Information Retrieval Music Genre Classification Hybrid Genetic Algorithm k-nearest neighbors Classifier multi-class classification
Unable to display preview. Download preview PDF.