Accelerated construction of stress relief music datasets using CNN and the Mel-scaled spectrogram

Area of Science:

• Computational Neuroscience
• Music Psychology
• Artificial Intelligence

Background:

• Music is a key tool for stress relief, but current options lack personalization.
• Traditional music curation for stress relies on costly, time-consuming biological measurements.
• Limited music choices can reduce the effectiveness of stress-relief interventions.

Purpose of the Study:

• To develop an efficient and economical deep learning approach for generating personalized stress-relief music.
• To overcome the limitations of traditional methods in curating music for stress reduction.
• To explore the potential of AI in enhancing music therapy and emotional well-being.

Main Methods:

• Utilized convolutional neural networks (CNNs) for music analysis.
• Generated large datasets of stress-relief music using Mel-scaled spectrograms.
• Extracted essential sound elements like frequency, amplitude, and waveform directly from music data.

Main Results:

• The deep learning model achieved a test accuracy of 98.7%.
• A clinical study confirmed the model-selected music's stress-relieving capacity matched researcher-verified selections.
• The method demonstrated efficiency and cost-effectiveness in creating music datasets.

Conclusions:

• Deep learning offers a transformative solution for personalized stress-relief music selection.
• The proposed method has significant implications for advancing music therapy through tailored interventions.
• This AI-driven approach enhances the potential for improved emotional well-being via personalized music.