Deep learning Binary/Multi classification for music's brainwave entrainment beats.

Area of Science:

• Neuroscience
• Music Psychology
• Artificial Intelligence

Background:

• Digital brain stimulant music, featuring entrainment beats, affects brainwaves similarly to medication.
• Previous research relied on electroencephalogram (EEG) signals, which is complex and prone to noise.
• A novel, EEG-free approach is needed for accurate analysis of music's brainwave effects.

Purpose of the Study:

• To develop a simple, accurate, and reliable method for categorizing digital brain stimulant music based on signal elements.
• To create deep learning models for real-time detection and classification of music beats and their brainwave influence.
• To establish new datasets for training and testing these models.

Main Methods:

• Utilized VGGish and YAMNET transfer deep learning models for binary classification of music beats.
• Developed modified BW-VGGish and BW-YAMNET models for multi-classification of music beat influence on five brainwaves.
• Generated the Brainwave Entrainment Beats (BWEB) and Brainwave Music Manipulation (BWMM) datasets.

Main Results:

• VGGish and YAMNET models achieved high accuracy (98.5% and 98.4%) in detecting music beats.
• Modified BW-VGGish and BW-YAMNET models demonstrated average accuracy of 94.5% in classifying music beat influence.
• YAMNET is recommended for mobile applications due to lower power consumption and latency.

Conclusions:

• The proposed EEG-free deep learning models offer a reliable alternative for analyzing digital brain stimulant music.
• The developed datasets and models facilitate further research into music's psychological and physiological effects.
• This work overcomes limitations of previous EEG-dependent methods, enabling broader application.