An Effective Machine Learning Approach for Music Genre Classification with Mel Spectrograms and KNN

The automatic classification of musical genre is extremely useful in the indexing of music. In the area of music information retrieval (MIR), categorizing music files based on their genre is commonly a subjective task performed by humans. Our study develops a classifier to predict the genre of audio files. GTZAN genre classification dataset was used to extract important features from audio files. We implemented a K-Nearest Neighbor (KNN) and have chosen a value of k as 2, 5, and 10. With KNN, the best performance was achieved with the value of k=10. The proposed feature extraction and classification models end up with better accuracy in musical genre classification.