Implementation of convolutional neural networks for music rhythm generation using EEG technology

Abstract

Music plays a crucial role in human life, and musicians often require extensive training to transcribe their ideas into musical output. Brain Machine Interfaces for music applications have the potential to allow musicians to directly translate music into musical performances. Recent advancements in extracting musical data from brain responses to auditory stimuli and reconstructing music stimuli from brain signals show significant potential for this technology. In this research, we developed two Deep Convolutional Neural Network models: a regressor to predict Power Spectral Density values in music stimuli from EEG data, and a classifier to classify brain signals into musical chords matching the chromatic scale in Western music theory. The input for analysis was a continuous stream of brain waves from participants listening to naturalistic music. The evaluation metrics showed that the models performed reasonably well. However, further investigation is needed to improve the model's performance. The regression model indicated that the model explained 44.08% of the variance in the test data. The Pearson correlation coefficient suggested a significant positive linear relationship between actual and predicted values. The classification model achieved an overall accuracy of around 52% and a test loss of 2.04, suggesting that the model was not generalizing well to the validation set. The confusion matrix showed that the model made inaccurate predictions for some classes. Overall, the evaluation metrics suggest that the models have performed reasonably well, but there is still room for improvement in the accuracy of the obtained results. The study suggested future research directions to improve the models' performance, such as refining the signal processing and feature extraction methods, investigating the effect of different EEG recording parameters, and exploring the use of other neuroimaging models. This study concluded that there is great potential for further work in this area, and continued research and development could lead to significant advancements in the field of music generation using EEG technology.