Statistical learning (SL) refers to the conscious or unconscious learning of statistical properties or regularities in our environment. The ability to extract statistical regularities is crucial for a wide variety of cognitive functions,
and is connected to our ability to anticipate and form expectations about the world. The auditory domain is a particularly fruitful ground to study this topic: SL contributes to speech acquisition,
and has also been shown to underlie several aspects of music perception. We aim to further elucidate auditory statistical learning in order to translate our findings to the health sector, in the form of cognitive assessment and diagnostic tools, and potentially as a means to help support cognitive function.
Memory and Music
Music shows some remarkable properties in the context of memory. Memory for music is spared from some forms of dementia and memory for melody, in particular, is long lasting and shows strong resilience towards interference. We aim to discover and explore memory phenomena unique to music and seek ways to use them in cognitive training and diagnosis of memory impairing
Music can also be used, of course, for educational purposes. In a recent collaboration, computational methods were used to classify which songs would be best used to bootstrap second language learning. For those learning English as a second language, it is common to use music as a support tool, however, many songs do not have easy to understand lyrics. This project aimed to automatically identify which songs, from a corpus spanning several different genres, contain the most clear
and discernable lyrics. To accomplish this goal, we examined which features of music, across several different genres, contributed most to making the lyrics sound intelligible. That is, the system identifies which songs are best-suited to assist those learning English as a second language.
Motion and Music
Successful rehabilitation from stroke requires time, effort, motivation, and close supervision. Guided by the latest research, as well as our own findings in music perception, gamification, and movement therapy, we develop technology that aims to support cognitive and motor rehabilitation. In our applications, we are particularly interested in addressing motivational issues,
and incorporate medically-informed dynamic live feedback for movement correction.
Music and Mental Health
It is now well documented that music can have a positive effect on people’s emotional states. For example, listening to carefully selected music can decrease anxiety and depression in certain clinical populations, as well as pre- and post-operative patients. We aim to incorporate the therapeutic aspects of music in our applications supporting mental health and well-being.
Deep learning approaches for music generation
The idea of automatically generating music has been around since the birth of computing. Recent advances in cloud computing, big data processing and
deep learning neural networks give us the tools to tackle challenges in this field such as generating music with long term
structure and coherence,
and generating music that fits video content, or that has a certain emotion.
Computational modelling of tonal perception and musical expectation
Just as with language, our minds are continuously forming implicit expectations and predictions when we listen to music. These expectations arise in part from a lifetime of listening to music. That is, simply by being exposed to music, our minds gradually form internal predictive models about how music sounds,
and these expectations guide our perception of music. At the Music Cognition lab, we work on building computational models able to learn music and form expectations in the way the human mind does. These AI approaches are used to simulate human music perception, so as to better understand how music works in the brain, including tonal perception, and the perception of phrase boundaries in music.