Popular music and the role of vocal melody in perceived emotion

Abstract

The voice plays a crucial role in expressing emotion in popular music. However, the importance of the voice in this context has not been systematically assessed. This study investigates the emotional effect of vocal features in popular music. In particular, it focuses on nonverbal characteristics, including vocal melody and rhythm. To determine the efficacy of these features, they are used to construct a computational Music Emotion Recognition (MER) system. The system is based on the circumplex model that expresses emotion in terms of arousal and valence. Two independent studies were used to develop the system. The first study established models for predicting arousal and valence based on a range of acoustical and nonverbal vocal features. The second study was used for independent validation of these models. Results show that features describing rhythmic qualities of the vocal line produce emotion models with a high level of generalizability. In particular these models reliably predict emotional valence, a well-known issue in existing Music Emotion Recognition systems.

Music Emotion Recognition: Toward new, robust standards in personalized and context-sensitive applications

“…In addition, MER has been collectively evaluated since 2007 in the Music Information Retrieval Evaluation eXchange (MIREX) 7 Audio Mood Classification task -a benchmark strategy to unify evaluation practices [25]. Since 2013, the Multimedia Evaluation Benchmark (MediaEval) 8 has produced several open datasets.…”

“…Traditional MER systems have been mainly inspired on supervised learning, relying on the existence of an annotated music emotion dataset. Indeed, the research community has openly dissected several issues of the MER field: Sturm [5] pointed out the deceptive simplicity of assembling emotion datasets from "ground truths" that are difficult to generate; Schedl et al [6] reported low statistical inter-rater agreement of perceived emotion annotations; Lange and Frieler [7] described generalized inconsistency of subjective ratings of emotional attributes in music; Juslin [2] remarked the generalized confusion of listeners between the concepts of perceived and induced emotions, possibly impacting annotation reliability; Beveridge and Knox [8] highlighted the difficulty of discovering acoustic features responsible for expressing or inducing emotions; Schuller [9] described a paradigm shift from the design of hand-crafted features to data-learned features which has also extended to MER, where the "black box" nature of machine learning models is even more problematic for model explainability [10]. In a nutshell, these issues must be addressed in order to improve the quality and significance of MER research.…”

Mainstream popular music research: a musical update

“…2017). In its current state, though, this research field provides hardly any concrete or differentiated information about emotional responses to popular music in general and mainstream popular music in particular (see also Beveridge and Knox 2018, p. 412), as it seems to be concerned mostly with methodological issues in connection with dataset choices or classification processes. Markus Schedl sees one fundamental reason for relevant scientific work remaining in a ‘fledgling state’ in that ‘involving users, which is an obvious necessity to build user-aware approaches, is time-consuming and hardly feasible on a large scale – at least not in academia’ (Schedl 2016, p. 2) 25…”

A Streamlined Encoder/decoder Architecture for Melody Extraction

“…Melody extraction is the task that aims to estimate the fundamental frequency (F0) of the dominant melody. Automatic melody extraction has been an active topic of research in the literature, since it has many important downstream applications in music analysis and retrieval [1][2][3][4].…”