Analyzing Music Genre Classification in Bengali Songs: A Comparative Study with the Marsyas Dataset Using Advanced Audio Feature Extraction Techniques

Music genre classification has been one of the very first challenges in audio sound processing. The ability to represent the features of a sound signal as numbers gave rise to many applications in Music Information Retrieval (MIR). Unfortunately, audio processing for Bengali speech has not reached the state-of-art level of other languages like English. Our key contributions lie in addressing the scarcity of resources and labeled data in Bengali audio processing, and in exploring advanced feature extraction methods to extract essential features including MFCC (Mel-Frequency Cepstral Coefficients), spectral centroid, and chroma frequency. This study explores machine learning algorithms for classifying ten genres in the Marsyas dataset and applies the same approach to categorize six genres in the Bengali Song dataset, comprising 1000 and 2004 tracks, respectively. Despite the scarcity of Bengali dataset, the outcomes show potential, indicating the chance to improve audio processing for languages that are not well-represented, thereby enriching the diversity of audio signal processing literature.