Long Short-Term Memory-Based Music Analysis System for Music Therapy.

Area of Science:

• Computational musicology
• Artificial intelligence in medicine
• Neuroscience and music therapy

Background:

• Music therapy utilizes music activities to stimulate the brain for therapeutic purposes.
• Artificial intelligence (AI) integration is revolutionizing music therapy's diagnostic, treatment, and evaluation processes.
• Innovating AI-driven music therapy methods is crucial for enhancing treatment accuracy and expanding medical applications.

Purpose of the Study:

• To propose an AI-based algorithm for generating and classifying multi-voice music data.
• To develop a Multi-Voice Music Generation (MVMG) system utilizing a long short-term memory (LSTM) model.
• To leverage AI for advancing music therapy techniques and applications.

Main Methods:

• An autoencoder model was employed for music feature extraction and representation into text sequences.
• A long short-term memory (LSTM)-based model was developed for music generation and classification.
• The MVMG system was evaluated using single-melody MIDI and Chinese classical music datasets.

Main Results:

• The autoencoder-based feature extractor achieved a maximum accuracy of 95.3%.
• The LSTM-based model demonstrated an average F1-score of 95.68% for music analysis.
• The proposed LSTM model significantly outperformed a Deep Neural Network (DNN)-based classification model.

Conclusions:

• The developed MVMG system effectively generates and classifies multi-voice music data using LSTM.
• AI-powered music therapy shows significant potential for improving diagnostic and treatment accuracy.
• This research provides a foundation for further development of AI applications in music therapy and the medical field.