Modeling listeners' emotional response to music

Area of Science:

• Computational Musicology
• Affective Computing
• Music Information Retrieval

Background:

• Music's ability to communicate emotions is well-researched.
• Previous studies focused on individual musical features (tempo, mode, articulation, timbre).
• Few studies modeled emotional continua using diverse audio-based musical features.

Purpose of the Study:

• To develop and evaluate a computational model for predicting emotional responses to music.
• To model continua of expressed emotions using multiple audio-based musical features.
• To assess the model's ability to generalize across different Western music genres.

Main Methods:

• A four-phase model construction: theoretical feature selection, empirical validation, feature selection, and overall evaluation.
• Utilized audio-based representations of musical features.
• Reviewed existing research on music, emotions, and feature extraction.
• Demonstrated phases using film soundtrack examples.

Main Results:

• The developed computational models explain a significant portion of listeners' self-reported emotions.
• The models demonstrate potential for generalization across various Western music genres.
• The model construction framework provides a systematic approach to feature selection and evaluation.

Conclusions:

• Computational models can effectively predict emotional responses to music.
• The approach shows promise for applications in music information retrieval and affective computing.
• Further research can explore applications in film scoring and personalized music experiences.