Using Deep Learning to Recognize Therapeutic Effects of Music Based on Emotions

• Published in: Sensors (Basel, Switzerland) Vol. 23; no. 2; p. 986
• Main Authors: Modran, Horia Alexandru, Chamunorwa, Tinashe, Ursuțiu, Doru, Samoilă, Cornel, Hedeșiu, Horia
• Format: Journal Article
• Language: English
• Published: Switzerland MDPI AG 14.01.2023; MDPI
• Subjects: Algorithms; Artificial intelligence; Classification; Datasets; Deep Learning; Emotions; Harmony (Music); Humans; Information retrieval; Listening; Machine Learning; Melody; Mental health; Music; Music therapy; Musical performances; neural networks; Neural Networks, Computer; python; Signal processing; Stress; deep learning; python; music therapy; artificial intelligence; neural networks
• ISSN: 1424-8220; 1424-8220
• DOI: 10.3390/s23020986

Music is important in everyday life, and music therapy can help treat a variety of health issues. Music listening is a technique used by music therapists in various clinical treatments. As a result, music therapists must have an intelligent system at their disposal to assist and support them in selecting the most appropriate music for each patient. Previous research has not thoroughly addressed the relationship between music features and their effects on patients. The current paper focuses on identifying and predicting whether music has therapeutic benefits. A machine learning model is developed, using a multi-class neural network to classify emotions into four categories and then predict the output. The neural network developed has three layers: (i) an input layer with multiple features; (ii) a deep connected hidden layer; (iii) an output layer. K-Fold Cross Validation was used to assess the estimator. The experiment aims to create a machine-learning model that can predict whether a specific song has therapeutic effects on a specific person. The model considers a person’s musical and emotional characteristics but is also trained to consider solfeggio frequencies. During the training phase, a subset of the Million Dataset is used. The user selects their favorite type of music and their current mood to allow the model to make a prediction. If the selected song is inappropriate, the application, using Machine Learning, recommends another type of music that may be useful for that specific user. An ongoing study is underway to validate the Machine Learning model. The developed system has been tested on many individuals. Because it achieved very good performance indicators, the proposed solution can be used by music therapists or even patients to select the appropriate song for their treatment.