Performance of WaveNet Variational Autoencoder Trained on Classical Piano Music

Date of Award

5-9-2026

Degree Name

M.C.S. in Computer Science

Department

Department of Computer Science

Advisor/Chair

Ju Shen

Abstract

Intelligent networks are rapidly advancing in their capabilities of producing seemingly authentic media, especially in areas related to text and image generation. In this paper, we propose an artificial intelligence-based model structure designed to produce authentic-sounding classical piano music. We combine the capabilities of WaveNet, a neural network created to autoregressively generate audio, and the variational autoencoder, a structure of model that learns to represent data in a continuous latent distribution. The model is trained on a dataset of classical piano music and then used to generate new synthetic music samples. During the generative process, we provide the model with different contexts that aid in informing its outputs. Our experimental results demonstrate that a WaveNet-based variational autoencoder is effective at modeling the data for digitized piano audio, but these results come up well short of passing as genuine piano audio. We conclude that this model structure has the potential to produce compelling audio data, but such results will likely require an extensive amount of training time to achieve.

Keywords

Artificial Intelligence, Computer Science

Comments

OCLC No. 1591829502

Rights Statement

Copyright 2026, author.

Link to Full Text

Share

COinS
 
 
 

Links