Music genre classification using MIDI and audio features
EURASIP Journal on Applied Signal Processing
Information shared by many objects
Proceedings of the 17th ACM conference on Information and knowledge management
IEEE Transactions on Information Theory
| Hi-index | 0.00 |
Normalized Compression Distance (NCD) is an information-theory based similarity metric that has been used successfully for similarity measuring in various domains, including music.. Here, we extend NCD from the pairwise similarity measurement to lists of objects. Based on the compressibility of a single object in the context of a list, we can make assumptions of the objects similarity with the objects in the given list. We apply the list compression into the task of classifying music in audio format according to the genre. We use features derived from a set of audio data as the list representations of a genre. Then, for a single piece of music that needs to be classified, we extract the same features and measure how well the representation compresses with a list of same features from different pieces of music. We present three variants of practical implementations of the. system. The evaluation shows that our approach has potential. Preliminary results seem promising and suggest that the method can be extended by applying more features into the measurement process. In addition, using different features allows utilizing the method for other classification tasks.