1:Hardware The basis for how this project functions overall is via an electroencephalogram dry scalp headset. An electroencephalogram(EEG) works via measuring the electrical potential between the difference in two points of voltage, in our case the ear in the scalp. For those of you unfamiliar with biology these voltages differences are due to electrical currents running in our brains neurons. These difference across action potential in the brain can be measured in order to measure the raw EEG signal. Now this EEG signal can be decomposed into its frequency components via a fast fourier transform performed on the Neuro Sky ASIC. These different frequency component are portions of the brain's signals which represent different activity and emotional levels. After these are transmitted over bluetooth to a UDP port which is connected to a laptop.
The basis behind the machine learning algorithm used to predict good sound predictions for future picks is signal processing methods. All of the signal processing for this project was performed in MATLAB. Basically, the start of the processing begins by reading in a CSV file containing the parsed brain wave data for Alpha, Beta, Delta, Theta and Gamma waves. After this is complete a data check is done to ensure the sensor was connected to the subject at all times by checking the connectivity variable. Once this is completed, the data obtained is normalized using a standard procedure of: x=x-x taking the data value subtracting the mean value and dividing by the standard deviation. The reason for the above is in order to even out all of the different variables for easier analysis and for easier machine learning applications. After completing the pre processing of the data the necessary signal features are constructed with the data. These features consist of the root mean square value,mean,variance and average absolute change of the brain wave signal. Mathematically the below definitions can be seen for each of the functions: Mean Equation:=1NXi Root Mean Square=1Nxi2Variance:21N(xi-)2 Average Amplitude Change:AAC=1Ny[n+1]-y[n]. Each of these mathematical factor help to distinction between the different genre and in addition the brain attention to this factor. After obtaining each of these factors a 20 Feature Vector by 20 Training Vector is developed. This is then fed into a neural network consisting of 10 nodes which can be able to determine the likeability index of the song based on likeability values for the training set. Note these values are only need for the infancy of the algorithm in order to be able to understand your brains patterns and better develop a model. After that point there is no longer a need for this model to rate since the sensor will be calibrated and have data for a variety of fluctuation periods.