SCS Undergraduate Thesis Topics

2012-2013
Student Advisor Thesis Topic
Dan Howarth Dr. Tai Sing Lee Predicting an individual's brain state induced by music listening

Our general goal is to establish a relationship between the features of a musical piece and the features of the neural activities induced by listening to a piece of music. This work provides evidence that this relationship exists and can be modelled. Furthermore, it provides evidence that a model learned on a number of song brain state pairs generalizes to predict the neural activities corresponding to unheard pieces of music. Modelling this relationship will allow us to create novel applications in music information retrieval and music therapy. Music information retrieval applications include the retrieval or synthesis of a piece of music to suit an individual's preference or desired cognitive state. One music therapy application is the retrieval of pieces of music that aid in priming the brain for particular tasks.

To this end, in this study we first recorded brain activities of multiple subjects through electroencephalography (EEG) while they repeatedly listened to two different pieces of music. Classification of the musical piece that the subject was listening to was performed in order to ensure that listening to a song induces repeatable brain states.

After this successful classification we ran another experiment in which a subject repeatedly listened to music from two classes. Within each class the songs are similar to each other in the music feature space? across classes they are dissimilar. In order to ensure our model can generalize to unheard pieces of music classification was performed with a cross validation scheme that consisted of training on the EEG examples corresponding to a fraction of the songs in each set and then classifying the examples corresponding to the remaining songs. We achieved classification accuracy above 90%. The success of this classification clearly shows that similar music induces similar brain states, and thus exhibits the feasibility of our general goal.


Close this window