The role of loudness in vocal intimidation.

Abstract

Across many species, a major function of vocal communication is to convey formidability, with low voice frequencies traditionally considered the main vehicle for projecting large size and aggression. Vocal loudness is often ignored, yet it might explain some puzzling exceptions to this frequency code. Here we demonstrate, through acoustic analyses of over 3,000 human vocalizations and four perceptual experiments, that vocalizers produce low frequencies when attempting to sound large, but loudness is prioritized for displays of strength and aggression. Our results show that, although being loud is effective for signaling strength and aggression, it poses a physiological trade-off with low frequencies because a loud voice is achieved by elevating pitch and opening the mouth wide into a-like vowels. This may explain why aggressive vocalizations are often high-pitched and why open vowels are considered "large" in sound symbolism despite their high first formant. Callers often compensate by adding vocal harshness (nonlinear vocal phenomena) to undesirably high-pitched loud vocalizations, but a combination of low and loud remains an honest predictor of both perceived and actual physical formidability. The proposed notion of a loudness-frequency trade-off thus adds a new dimension to the widely accepted frequency code and requires a fundamental rethinking of the evolutionary forces shaping the form of acoustic signals. Public Significance StatementSpeakers can intimidate by being loud or by lowering voice frequency: low frequency appears to work best for size exaggeration, whereas loudness is more effective for displaying strength and aggression. However, it is physiologically difficult to achieve both simultaneously, making the low-and-loud combination an honest index of physical formidability and the elusive key to vocal intimidation.

NLP and speech intelligibility

“…This has important implications for other forms of vocal communication besides speech. Angry roars and screams have a high f o as a byproduct of beingloud [34], but the tradeoff is that their spectrum has too few harmonics to excite vocal tract resonances (see e.g., Fig 1A ), which means that they may fail to encode filter-related formant information and potentially become less individually distinct [56,57]. If NLP partly compensate for this high f o and keep formants audible, while also lowering the apparent pitch and avoiding a "tinny" voice quality [44], they may be highly adaptive for vocal intimidation [34].…”

“…Angry roars and screams have a high f o as a byproduct of beingloud [34], but the tradeoff is that their spectrum has too few harmonics to excite vocal tract resonances (see e.g., Fig 1A ), which means that they may fail to encode filter-related formant information and potentially become less individually distinct [56,57]. If NLP partly compensate for this high f o and keep formants audible, while also lowering the apparent pitch and avoiding a "tinny" voice quality [44], they may be highly adaptive for vocal intimidation [34]. While we did not directly investigate shouted speech or singing, we have shown in past work that shouted speech gives rise to more NLP and is an effective strategy for vocal intimidation [34] It is a reasonable hypothesis to test in followup studies that the intelligibility of shouted speech at high f o can be enhanced by volitionally adding some vocal roughness -for example, by engaging the ventricular folds or the arytenoids, as in rock singing [41].…”

“…1E). At the highest tested pitch of 880 Hz, diphthongs with versus without NLP had much lower discrimination thresholds: AM by 14% [8,21], subharmonics 18% [12,25], chaos 23% [17,30], and whisper 28% [22,34]. Interestingly, AM and subharmonics did not cause a noticeable "diphthong bias" at very high f o (only +2% [0, 4] and 6% [4,10], respectively, at f o = 880 Hz and morph = 0), whereas the whisper and chaos manipulations were more likely to create a misleading perception of change in a pure vowel (+16% [11,22] and 23% [17,30], respectively; see Fig.…”

“…An interesting alternative approach to searching for acoustic changes that degrade speech intelligibility is to ask how speech can be made easier to understand. The speaking style known as clear speech tends to have a more variable and higher f o with more energy in high frequencies (possibly as a side effect of speaking more loudly [34]), longer phonemes and pauses, clearer syllable articulation, emphasized consonants, an expanded vowel space, clear formant transitions (e.g., in voiced stops), narrower formant bandwidths, and fewer vowel reductions [24]. These characteristics are consistent with the notion that formants are central to speech comprehension.…”

How to analyze and manipulate nonlinear phenomena in voice recordings

“…Listeners may also possess sensory biases, in which case the effect of a perceptual feature may be partly decoupled from the biological information it encodes. For example, humans and many nonhuman animals strongly associate low frequencies [8], high intensity [9], and acoustic roughness [10] with size and formidability, which callers can exploit to achieve acoustic size exaggeration. Ideally, this requires models linking objective acoustic properties with percepts, but human psychoacoustics is a long way from achieving comprehensive perceptual models, and even less is known about the perception of NLP in nonhuman animals.…”