Student Members: Garrett Archer, Hiromi Kageyama
Faculty Advisors: Dr. Blair Kaneshiro, Dr. Masoumeh Heidari
Description:
The ability to accurately predict song tempo from a subject’s electroencephalogram (EEG) data
plays a crucial role in understanding music perception and cognition. This project set out to
determine if this data could be used for accurate prediction without using machine learning
methods, which require training and testing data that may not be available, and which frequency
range is the most effective for this prediction. We used data recorded from 20 participants who
listened to a selection of 10 songs. We created a program in MatLab that optimized this data for
every song and plotted the most significant peaks for each of the three ranges that we had
chosen on a categorical scatter plot. The results were then compared to the actual tempo of
that song. For most songs we found this method to be quite accurate, especially in the 3-7 Hz
range, though a couple songs proved to have more difficult tempos to predict. This program not
only can find use in its current state for determining tempo from pre-recorded data, but future
developments could see it being adapted for real-time tempo estimation or for use in
EEG-based music composition.
Comments are closed