Dissertation/Thesis Abstract

DebuSequencer: Applying Statistical Sequence Prediction Techniques to Music Composition Using Generative Adversarial Networks
by Battalino, Jesse, M.A., Mills College, 2019, 100; 13900143
Abstract (Summary)

Sequence generation can be a difficult problem to solve using strictly heuristic approaches. To decide on the next value in a series of values requires a level of understanding that must be unique to a given series, and this is challenging to represent in a finite feature space. Using Machine Learning principles, researchers have developed strategies for modeling sequence prediction behavior, which can be applied to a wide range of data sets including shopping trends, predictive text, and music composition. As with any potential solution to a problem that utilizes Machine Learning techniques, deciding which features of input values are statically bound to desirable output is only one of many concerns. In the case of music composition, elements of music such as phrase, key, tempo, melody, accompaniment, and various forms of modulation must also play a part in giving a piece its overall quality and allows the listener to perceive its musical language. This project, DebuSequencer, will explore the use of pitch, octave, dynamics, tempo, and key from MIDI instruction data as features to train a pair of Recurrent Neural Networks connected as a Generative Adversarial Network. The methods of feature selection and network configuration formulated within this project will generate sequences in the style of various types of music.

Indexing (document details)
Advisor: Wang, Susan
Commitee: Bischoff, John L., Oparah, Chinyere
School: Mills College
Department: Interdisciplinary Computer Science
School Location: United States -- California
Source: MAI 81/2(E), Masters Abstracts International
Source Type: DISSERTATION
Subjects: Artificial intelligence, Computer science, Musical composition
Keywords: GAN, Keras, LSTM, Machine learning, Music composition, RNN
Publication Number: 13900143
ISBN: 9781085611404
Copyright © 2019 ProQuest LLC. All rights reserved. Terms and Conditions Privacy Policy Cookie Policy
ProQuest