The Classical-Romantic Dichotomy: A Machine Learning Approach

Abstract

Amidst the rise of audio media subscription services, the demand for automated Music Information Retrieval (MIR) and musical classification grows. Traditional methods, like the Mel-Frequency Cepstral Coefficients (MFCC), focus on low-level sound details. This study turns to Musical Instrument Digital Interface (MIDI) files to tap into mid-level musical nuances. Using MATLAB’s MIDI-Tools package, we aimed to classify between Classical and Romantic music using Support Vector Machines (SVMs) and Long Short Term Memory (LSTM) networks. Despite achieving 82% accuracy with LSTM and 72% with SVM, the model’s extensibility to more categories remains unproven. Incorporating advancements from Convolutional Neural Networks might enhance the model. A fusion of MIDI-based mid-level details with sound spectrum analysis and SVMs could further improve future classifiers.