Multi-Modality in Music: Predicting Emotion in Music from High-Level Audio Features and Lyrics

TL;DR

This study investigates whether combining high-level audio features and lyrics improves music emotion recognition, finding that multi-modal approaches outperform uni-modal audio in predicting valence, with specific features contributing most.

Contribution

It demonstrates that multi-modal features enhance emotion prediction in music over audio-only methods, identifying key features that contribute to improved performance.

Findings

Multi-modal features outperform audio-only in valence prediction.

Five key high-level features significantly contribute to performance.

Code for the approach is publicly available.

Abstract

This paper aims to test whether a multi-modal approach for music emotion recognition (MER) performs better than a uni-modal one on high-level song features and lyrics. We use 11 song features retrieved from the Spotify API, combined lyrics features including sentiment, TF-IDF, and Anew to predict valence and arousal (Russell, 1980) scores on the Deezer Mood Detection Dataset (DMDD) (Delbouys et al., 2018) with 4 different regression models. We find that out of the 11 high-level song features, mainly 5 contribute to the performance, multi-modal features do better than audio alone when predicting valence. We made our code publically available.