Discriminating real from machine-learning generated music

People

Supervisor

Research areas

A stylised neural network classifying a musical excerpt

Description

Most generative music systems just create new music, this project would involve training discriminator neural networks to tell if music was created by a human, or a cutting edge ANN such as Music Transformer or PerformanceRNN.

To accomplish this project, you will learn about current trends in music generation and machine learning using machine learning. You will create your own neural network or other ML system and train it on generated and real musical examples. You could choose to work either in the symbolic music domain (i.e., with MIDI files) or with real digital audio.

Goals

  • gain an understanding of current trends in machine listening and computer music generation
  • design a neural network or other ML system to discriminate real from generated music
  • test this system with music from cutting edge music generation systems

Requirements

  • skills in Python programming
  • completed coursework in machine learning or artificial intelligence
  • understanding of basic concepts of music computing
  • interest and motivation to work with music

Keywords

music generation, machine learning, music information retrieval, computer listening

Updated:  1 June 2019/Responsible Officer:  Dean, CECS/Page Contact:  CECS Marketing