Detecting Digital Stimulant music Using Bidirectional Deep Long Short Term Memory

Abstract

Recently, researchers become interested in discovering patterns in brain activity that correspond to different emotions, so the link between musical stimuli and brain waves has received a lot of attention. Music with brainwave entrainment beats (Digital Stimulant) is one of many factors that affect brain waves. Digital stimulants can result in positive effects on human neurons. Stimulants may be used as a therapeutic stimulus, such as pain reduction. Also, it can have negative effects on the brain when its use increases feelings of depression or provokes seizures in patients with epilepsy. All of the research on the effect of music on brain waves has been done by classifying electroencephalograms (EEGs) collected from volunteers while listening to various types of music. However, EEG-based recognition suffers from the difficulty of obtaining each piece of music and obtaining humans to test it. This leads to the proposed idea of using music features such as Mel-Frequency Cepstral Coefficients (MFCC) and Gammatone Cepstral Coefficients (GTCC) to classify music and determine whether or not it will affect the brain. Hence, this paper proposes an approach for music classification based on the presence or absence of digital stimulant based on bidirectional deep long short term memory (BDLSTM) architecture. Since, as far as we know, there is no classified dataset for the music files based on whether the brainwave entrainment beats exist or not, the paper also proposes a new brainwave entrainment beats (BWEB) dataset to evaluate the proposed model performance. Classification results showed that the best performance is obtained when using the bidirectional deep long short term memory (BDLSTM) as it achieved an accuracy of 90.2%, while deeper long short term memory achieved an accuracy of 88.4%.