The perception of tension and release dynamics constitutes one of the essential aspects of music listening. However, modeling musical tension to predict perception of listeners has been a challenge to researchers. Seminal work demonstrated that tension reported continuously by listeners can be accurately predicted from a discrete set of musical features, combining them into a weighted sum of slopes reflecting their combined dynamics over time. However, this model lacks an automatic pipeline for

more »

... feature extraction that would make it widely accessible to researchers in the field. Here, we propose an updated version of a predictive tension model that operates using a musical audio as the only input. Using state-of-the-art music information retrieval (MIR) methods, we automatically extract a set of five features (i.e., loudness, pitch height, dissonance, tempo, and onset frequency) to use as predictors for musical tension. The algorithm was trained to best predict behavioral tension ratings collected on a variety of pieces, and its performance was tested by assessing the correlation between the predicted tension and unseen continuous behavioral tension ratings. We hope that providing the research community with an open-source algorithm for predicting musical tension will motivate further work in the music cognition field for elucidating tension dynamics and its neural and cognitive correlates for various musical genres and cultures.