Generation of Music With Dynamics Using Deep Convolutional Generative Adversarial Network
Raymond Kwan How Toh, Alexei Sourin · 2021
Following the rapid advancement of Artificial Intelligence and transition into the era of Big Data, researchers have started to explore the possibility of using machine learning in creative domains such as music generation. However, most research were focused on musical composition and removed expressive attributes during data pre-processing, which resulted in mechanical-sounding generated music. To address this issue, music elements, such as pitch, time and velocity, were extracted from MIDI tracks and encoded with piano-roll data representation. With the piano-roll data representation, Deep Convolutional Generative Adversarial Network (DCGAN) learned the data distribution from the given dataset and generated new data derived from the same distribution. The generated music was evaluated based on its incorporation of music dynamics and a user study. The evaluation results verified that DCGAN could generate expressive music comprising of music dynamics and syncopated rhythm.