songₛentₛcores: Computational Design for Charting Dynamic Emotion in Songs with a Multimodal Circumplex Framework

AbstractMusic profoundly impacts human emotion, yet computationally modeling its dynamic affective qualities remains challenging. This pictorial introduces songₛentₛcores, a novel open-source toolkit (R/Python) that operationalizes Russell’s Circumplex Model of Affect to represent and visualize the dynamic interplay of Valence (pleasantness/unpleasantness) and Arousal (energy/calmness) in songs. The toolkit derives these dimensions multimodally from both the audio signal (via a CLAP model) and lyrical content (transcribed by ASR and analyzed with an NLI-based model). We visually demonstrate songₛentₛcores' capabilities through: (1) a small-scale experiment comparing its affective classifications against human ratings, revealing varied agreement across songs (e. g. , r = 0. 79 for ‘Negro Drama’, r = -0. 70 for ‘My Baby Just Cares for Me’) ; and (2) a case study analyzing ‘Negro Drama’ via its circumplex trajectory and an autoregressive affective network. This computational designapproach offers a richer, continuous representation of musical affect, providing a novel lens for designers, music therapists, and researchers to explore music’s emotional architecture. KeywordsMusic, Artificial Intelligence, Psychometrics, Computational Design, Circumplex Mode