Poster or Presentation Title
Advisor(s)
Mr. Aung Aung Phyo Wai
Mr. Claude Chua
Subjects
Computer Science
Abstract
Music is observed to possess significant beneficial effects to human mental health, especially for patients undergoing therapy and older adults. Prior research focusing on machine recognition of the emotion music induces by classifying low-level music features has utilized subjective annotation to label data for classification. We validate this approach by using an electroencephalography-based approach to cross-check the predictions of music emotion made with the predictions from low-level music feature data as well as collected subjective annotation data. Collecting 8-channel EEG data from 10 participants listening to segments of 40 songs from 5 different genres, we obtain a subject-independent classification accuracy for EEG test data of 98.2298% using an ensemble classifier. We also classify low-level music features to cross-check music emotion predictions from music features with the predictions from EEG data, obtaining a classification accuracy of 94.9774% using an ensemble classifier. We establish links between specific genre preference and perceived valence, validating individualized approaches towards music therapy. We then use the classification predictions from the EEG data and combine it with the predictions from music feature data and subjective annotations, showing the similarity of the predictions made by these approaches, validating an integrated approach with music features and subjective annotation to classify music emotion. We use the music feature-based approach to classify 250 popular songs from 5 genres and create a musical playlist application to create playlists based on existing psychological theory to contribute emotional benefit to individuals, validating our playlist methodology as an effective method to induce positive emotional response.
Included in
Artificial Intelligence and Robotics Commons, Biomedical Engineering and Bioengineering Commons, Graphics and Human Computer Interfaces Commons, Psychology Commons, Software Engineering Commons
Using EEG-validated Music Emotion Recognition Techniques to Classify Multi-Genre Popular Music for Therapeutic Purposes
Music is observed to possess significant beneficial effects to human mental health, especially for patients undergoing therapy and older adults. Prior research focusing on machine recognition of the emotion music induces by classifying low-level music features has utilized subjective annotation to label data for classification. We validate this approach by using an electroencephalography-based approach to cross-check the predictions of music emotion made with the predictions from low-level music feature data as well as collected subjective annotation data. Collecting 8-channel EEG data from 10 participants listening to segments of 40 songs from 5 different genres, we obtain a subject-independent classification accuracy for EEG test data of 98.2298% using an ensemble classifier. We also classify low-level music features to cross-check music emotion predictions from music features with the predictions from EEG data, obtaining a classification accuracy of 94.9774% using an ensemble classifier. We establish links between specific genre preference and perceived valence, validating individualized approaches towards music therapy. We then use the classification predictions from the EEG data and combine it with the predictions from music feature data and subjective annotations, showing the similarity of the predictions made by these approaches, validating an integrated approach with music features and subjective annotation to classify music emotion. We use the music feature-based approach to classify 250 popular songs from 5 genres and create a musical playlist application to create playlists based on existing psychological theory to contribute emotional benefit to individuals, validating our playlist methodology as an effective method to induce positive emotional response.