Browsing by Author "Arriagada Stoller, Pascal"
Now showing 1 - 1 of 1
- Results Per Page
- Sort Options
Thesis A Transformer Approach for the Analysis of Music’s Emotional Trajectory Directly from Music Audio for Recommender System Applications(2024-03) Arriagada Stoller, Pascal; Araya López, Mauricio Alejandro (Profesor Guía); Zañartu Salas, Matías (Profesor Correferente); Moreno Araya, Sebastián (Profesor Correferente); Universidad Técnica Federico Santa María. Departamento de Ingeniería EléctricaUnder the research field of Music Dynamic Emotion Recognition, a subset of Music Information Retrieval and Affective Computing, this study delves into research on Deep Learning techniques focused on the estimation of values associated with the perceived emotional trajectory of complete musical pieces or excerpts. Based on this, one of the research objectives consisted of the implementation of a transformer model that provides dynamic predictions, which are then used for an emotion-based Musical Recommendation System. The concept of emotional trajectory is defined under Russell’s Circumplex model of Affect (1980), a dynamic emotional theory based on the orthogonal dimensions of Arousal and Valence. The models to implement utilize as input a set of 260 low-level features extracted via the openSMILE toolkit. These features were selected during the generation process of the MediaEval dataset for Emotional Analysis in Music (2015), which comprises 1802 music pieces along with dynamic emotional annotations. This dataset serves as a benchmark in the field, offering selected metrics to assess model performance and predictive behavior, obtaining results similar to the state-of-the-art, thus demonstrating the applicability of this type of user-oriented systems.
