Proposing scalable method for music genre classification

Music, a vital part of our lives from long time. It is a one of the type of multimedia and much online music content distribution vendors are providing very large database of musical libraries for streaming and downloading services. Classification of audio files by genres have a significant role in management of music. Music genre classification helps in organizing, searching and retrieving musical content. It is also used for music recommendation system. In this paper, music genre classification using serial execution and parallel execution is implemented. For both serial and parallel execution, feature vector is extracted from mel-frequency cepstral coefficients (MFCC) and some other features and support vector machine (SVM) is used for genre classification. The classification accuracy of the proposed single-node system is 88.40%. The paper mainly focuses on multi-node system and how it can be useful to improve performance of large musical database. Apache Spark enables fast processing of data in hadoop clusters. It also allows to persist data in memory. Spark have support of machine learning, graph processing and SQL queries.