Files
Abstract
This work explores how humans use elements of prosody (e.g., pitch, loudness) to communicate social identity and emotional affect. Understanding these components of language is crucial to effectively responding (e.g., the word ‘yes’ said excitedly warrants a different response than the same word said angrily); however, the prosodic cues involved in communicating such elements are understudied, particularly in natural speech. A qualitative examination of archival speech corpora of naturalistic personal interviews for socially diverse speech (Digital Archive of Southern Speech), naturalistic emotional speech (StoryCorps), and acted emotional speech (RAVDESS) was undertaken, followed by acoustic measurement and quantitative analysis of the following prosodic cues in these samples: f0 trajectory shape and height, mean f0, f0 range, mean intensity, intensity range, vowel duration, jitter, shimmer, and harmonics-to-noise ratio.
The methodology utilized here is a departure from much previous work in that it examines these acoustic prosodic cues in individual vowel segments as opposed to larger phrase-level prosodic utterances. In doing so, this work employs automated methods for vowel identification and acoustic measurement that can be readily applied to larger and more expansive corpora in future studies. Furthermore, generalized additive mixed modeling is utilized to analyze dynamic aspects of prosody (e.g., the shape of f0 trajectory over the time course of each vowel), and linear mixed-effects regression modeling is employed to investigate the impact of social identity and emotional affect on static prosodic elements.
Results indicate that factors of social identity such as sex, class, ethnicity, region, and age are significant predictors of pitch trajectory shape. In terms of emotional affect, trajectory shape varies by level of arousal (low activation to high activation) and valence (negative to positive). Furthermore, static acoustic cues such as mean pitch and range also vary based on these social and emotional factors. By establishing methods for investigating prosody on a large scale in an efficient and fruitful way, this study contributes to our understanding of how we express who we are and how we feel in spoken language.
The methodology utilized here is a departure from much previous work in that it examines these acoustic prosodic cues in individual vowel segments as opposed to larger phrase-level prosodic utterances. In doing so, this work employs automated methods for vowel identification and acoustic measurement that can be readily applied to larger and more expansive corpora in future studies. Furthermore, generalized additive mixed modeling is utilized to analyze dynamic aspects of prosody (e.g., the shape of f0 trajectory over the time course of each vowel), and linear mixed-effects regression modeling is employed to investigate the impact of social identity and emotional affect on static prosodic elements.
Results indicate that factors of social identity such as sex, class, ethnicity, region, and age are significant predictors of pitch trajectory shape. In terms of emotional affect, trajectory shape varies by level of arousal (low activation to high activation) and valence (negative to positive). Furthermore, static acoustic cues such as mean pitch and range also vary based on these social and emotional factors. By establishing methods for investigating prosody on a large scale in an efficient and fruitful way, this study contributes to our understanding of how we express who we are and how we feel in spoken language.