Web-Based Music Genre Classification for Timeline Song Visualization and Analysis

Abstract

This paper describes a web application that retrieves songs from YouTube and categorises
them. The tool described in this study is based on models trained using Audioset's musical
collection data. We used classifiers from various Machine Learning paradigms for this
purpose, including Probabilistic Graphical Models (Naive Bayes), Feed-forward and
Recurrent Neural Networks, and Support Vector Machines (SVMs). In a multi-label
classification scenario, all of these models were trained. Because genres can change over the
course of a song, we classify it in ten-second increments. Audioset, which provides 10-
second samples, enables this capability. The visualisation output displays this temporal
information in real time, in sync with the music video being played, with classification results
displayed in stacked area charts, with scores for the top-10 labels obtained per chunk shown.
We briefly explain the problem's theoretical and scientific foundations, as well as the
proposed classifiers.