Font Size: 

In recent years, advances in AI have led to a rapid increase in tools such as natural languagegeneration and image generation. AI technologies in the music field include Izotope'sOzone11 for automatic mastering and Mubert's tool for automatically generating music fromtext. However, AI-based music analysis tools for classification purposes are not yetwidespread.This research aims to use AI and machine learning to classify music genres and assist artists.By objectively analyzing their own music, artists can better understand the characteristics oftheir music and incorporate popular trends. Additionally, they can also evaluate their ownwork, set goals, and drive continuous improvement. In this study, we use Music2vec[1] as amethod to convert music into vector representations for classification. We also exploreprevious work using Soundnet[2] and CRNN for music analysis. The definition of musicalgenre establishes its two defining criteria: classification based on performance form andmusical structure and classification based on historical description.In this study, we conducted experiments using a dataset of UK Hardcore and J-core genres totrain machine learning models. The inputs consisted of 30-second tracks, which were used asMel spectrograms. We trained the models on 180 songs from each genre and compared of themodels with and without stride. The model trained without stride achieved 71% accuracy,while the model trained with stride achieved 80% accuracy. This shows that using stride toexpand the dataset can lead to higher accuracy.This research also demonstrates new possibilities for artists, showing how AI can provideinsights into their music that they may not have been aware of. By using these insights in selfpromotion, artists can reach an audience that more closely matches with their intended market.

Music2vec, Mel Spectrogram, Soundnet