Real-Time Human-Classified Emotional MIDI Dataset Integration for Symbolic Music Generation

This paper explores the interplay between affect and musical engagement, drawing on psychological theories and empirical studies. We propose extending this exploration into telematic music, investigating the potential for creating an emotional AI collaborator. The study aims to compile a dataset for machine learning by converting physiological re-sponses, self-reported emotions, and musical passages. That data will allow for the generation of symbolic music from AI. The relationship between affect (emotional experience) and musical engagement is explored across psychology, neuroscience, and musicology, offering insights into how music evokes emotions and engages individuals cognitively. This exploration involves a social-psychological inquiry into human perceptions and feelings towards musical engagement and a deeper investigation into the physiological and neuroscientific responses when exposed to musical stimuli. The project aims to create a new MIDI dataset with real-time emotional data captured from L20rk Tweeter, a telematic musicking software platform, and integrate it into the symbolic music generation process. This involves the development of a module for real-time emotion classification and incorporation into the symbolic music generation pipeline. Integrating AI, machine learning, and human feedback is proposed to deepen understanding and enable continuous measurement of training data, thereby advancing music information retrieval knowledge and enhancing emotional expression, collaboration, and engage-ment in musical performance.