Currently Recruiting! I’m currently looking for PhD students in the field of musical machine learning, particularly for projects focussed on using AI/ML in live performance!
If you’re passionate about music, machine learning, and computer science (or some weighting of those three fields), and want to be a part of an innovative college at one of the world’s top universities, get in touch through ANU, or on Twitter!
I currently teach:
- Laptop Ensemble (LENS) at the Australian National University (S2 2019, S1+2 2020)
- Computer Organisation and Program Execution (COMP2300) at the Australian National University (S1 2019, S1 2020).
I used to teach:
- Advanced Topics in AI for Intelligent Systems (IN5490) at the University of Oslo,
- Art and Interaction in New Media (COMP1720),
- Drums and Percussion at the ANU School of Music.
I supervise and co-supervise PhD and Master’s students at the ANU and at University of Oslo. My previous students’ published work is listed here.
PhD Students (co-supervisor):
- Tønnes Nygaard (http://robotikk.net/) is studying evolutionary robotics at the University of Oslo, Department of Informatics and has created a open-source quadruped robotics platform with mechanically extensible legs for evolving control systems and robot morphology simultaneously during real-world activity. (2018-2020)
- Benedikte Wallace (https://www.hf.uio.no/ritmo/english/people/phd-fellows/benediwa/) is studying machine learning models of sound-related movement and dance, among other creative applications of artificial intelligence, at the RITMO Centre of Excellence, University of Oslo. (2018-2021)
I’m available to supervise student projects in sound and music computing, creativity support systems, computational creativity, music technology, and interactive systems. I’m interested in creating and studying new computing systems at the nexus of creative arts, physical embodiment, interaction, and intelligence.
Projects could involve:
- developing predictive musical instruments
- machine learning of musical style
- musical AI
- computer support for collaborative musical expression
- new interfaces for musical expression (NIME)
- applying ML/AI in creative practices
Here’s some project ideas that could be extended or shaped to suit you:
Discriminating real from neural generated music
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.
Generating Harmony from Melody
This project involves creating a sequence-to-sequence ANN that can generate harmony, or counter-melodic parts for a given melody. The focus here would be to deploy this network in a music creating app such as MicroJam.
Generating “improved” versions of a melody/MIDI content
The idea of this project is to take an existing melody created in an interactive music system and change its style, or improve it in some way. It could involve modifying the melody to fit a particular harmony, or create a more appealing melodic shape. Various sequence-to-sequence ANN techniques could be applied here, and such a system could have applications in musical performance and education.
Neural Networks for Generating Digital Audio
The idea of generating digital audio directly from a neural network is exciting and there are stunning results from examples such as Wavenet, SampleRNN, FFTNet, and WaveGan. Projects in this area will involve finding new application areas for neural audio generation and developing focussed and fast algorithms with appropriate training data for use in these areas. In particular, the use of these networks for creative arts is still being actively explored.
Evolutionary Music Making
Making music with evolutionary algorithms has a long history, but is yet to break into mainstream music technology systems. In this project, you will develop an interactive music application (e.g., web or mobile app) for generating and assessing music created using an evolutionary algorithm. A novel approach might involve the human user evaluating some generated sounds, while others are analysed by an automatic system using MIR techniques or a discriminator neural network.
New Applications of the Mixture Density Recurrent Neural Network
The MDRNN is an exciting sequence model that can generate multiple continuous valued signals from a Gaussian mixture model at each step in time. This project will involve applying and extending my Keras MDRNN models into new applications in the creative arts and beyond. We’ve tried using the MDRNN for voice synthesis, motion capture data synthesis, and musical control data synthesis, but there are lots of other potential applications waiting for you to discover, e.g.: predicting future sensor values, generating robot movements, generating world models for video games or real life situations etc.