In its more than 60 year history, music generation systems have never been more popular than today. In this talk, Prof. Dorien Herremans will give a brief history of the unique field of algorithmic composition. While there are a number of systems out there, two main challenges remain: how do we integrate emotion as well as ensure long term structure in musical pieces. It should come at no surprise that music and emotion are intrinsically connected. Yet, computer systems still struggle to truly capture emotion. Secondly, while many systems can compose short fragments that sound well, the real challenge is to create complete, longer pieces of music with patterns and themes. The MorpheuS system, a hybrid machine learning system developed by Prof. Dorien during her MSCA fellowship with Prof. Elaine Chew, tackles these two challenges. In this talk we will dive into the inner workings of MorpheuS, as well as listen to a number of generated fragments performed by Prof. Elaine Chew. Finally, we will discuss some recent advances in the field and what respective labs are currently working on.
What is this?
‘Meet my Lab’ is a virtual meeting point profiling researchers and their work. The presenters will share their research work and present opportunities for collaboration. The focus is on interactivity - audience members can ask questions and engage with the presenter(s) and with each other.
Who can participate?
Meet my Lab’ is explicitly open to the world! We welcome researchers at all career stages and of all nationalities.
Participants should have an interest in international research collaboration!
Participation is free of charge. Seats will be allocated on a first come first served basis.
***
Short synopsis of respective work and collaborative work
Dorien Herremans' recent work has focused on creating controllable music generation systems using deep learning technologies. One challenge in particular - generating music with steerable emotion - has been central in her research. When it comes to affect and emotion, computer models still do not compare to humans. Using affective computing techniques and deep learning, Dorien's team has built models that learn to predict perceived emotion from music. These models are then used to generate new fragments in a controllable manner, so that users can steer the desired arousal/valence level or tension in newly generated music. Other challenges tackled by Dorien's team include ensuring repeated themes in music, automatic music transcription, and novel music representations.
Elaine Chew is known for inventing the spiral array model, a geometric representation for tonality. Her research focuses on the mathematical and computational modelling of musical structures in music, with special interest in expressive performance, and electrocardiographic sequences. Applications include modelling of music performance, AI music generation, music-heart-brain interactions, and computational arrhythmia research. She is author of numerous academic publications and centre of one of 9 publication clusters having ≥5 women in the international Music Information Retrieval community (ISMIR 2016 infometric study). As a pianist, she integrates her research into concert-conversations that showcase scientific visualisations and lab-grown compositions.
Together, Herremans and Chew worked on the MorpheuS automatic music composition system. To guide MorpheuS' music generation, they designed and implemented a harmonic tension model based on the spiral array. The system learns rhythms and harmonic tension profiles from a template piece to generate music with long term structure. MorpheuS' creations have been featured on Channel News Asia's television program on "Algorithms: Part 1 - Rage Against The Machine", where it was pitted against a human composer and had its output (based on the first of Three Pieces for String Quartet by Stravinsky) performed by members of the Singapore Symphony Orchestra. Chew has also performed MorpheuS pieces at concerts in venues in Cambridge (MA), London, San Francisco, and Stanford.
