Estimation of musical features using EEG signals

Nowadays with the recent development in Brain Computer Interfaces (BCI), research field branches to arts and especially to music. In our study, a system is developed which analyses and classifies musical features from Fp1 region of the subjects' frontal region of the scalp and estimates musical characteristics using deep learning methods. In order to determine the musical characteristics that is going to be estimated, various musical analyzation techniques are carried out. Acquisition conducted using a single channel dry electrode EEG device. Artifacts such as low frequency of eye blinking and noise around the electrodes are supressed from EEG signals using preprocessing. Spectrogram matrices were created from EEG signals and feed as inputs to deep learning models. It has been observed that the LSTM layer added to the convolutional neural networks has achieved high accuracy in the classification of musical characteristic that were extracted using analyzing of the EEG signals.