Effects of stimulus amplitude-scaling approach on emotional responses to non-speech sounds

Abstract

In the study of auditory emotion perception, it is important to calibrate test sounds so their presentation level during testing is known. It is also often desirable to standardize the amplitude of the sounds so that each sound used in testing is approximately the same level. However, existing literature in the study of auditory emotion perception includes a mixture of techniques for standardizing amplitude across sounds. The purpose of this study was to compare the effects of two amplitude-scaling approaches on emotional responses to non-speech sounds, specifically standardization based on peak level or root-mean-square (rms) level. Nineteen young adults provided ratings of valence and arousal via an online testing program. Stimuli were non-speech sounds scaled in two ways, based on the stimulus' peak level or rms level. Ratings were analyzed using linear-mixed effects modeling to compare scaling methods; correlations between ratings and level within each scaling method were explored. Analysis revealed that the ratings of peak-scaled sounds were less pleasant and more exciting than were the ratings of rms-scaled sounds, although the effects were small in magnitude (~0.2 points on a 1-9 scale). Within rms-scaled sounds, peak level was not related to ratings of valence or arousal. However, within peak-scaled sounds, rms level was related to ratings of valence and arousal. Combined, these data suggest that amplitude standardization has a small effect on ratings overall, but investigators might be motivated to choose one approach over the other, depending on the research question. Rms-scaling reduces overall level as a cue for emotional responses, while peak-scaling maintains some natural variability in responses related to level. Finally, results are specific to this stimulus set. The effects of amplitude-scaling would be expected to be negligible for a stimulus set where the sounds have homogenous temporal dynamics.

Fig 1. Examples of (A) rms- and (B) peak-scaled sound samples.

The figure visually highlights differences between methodologies. Within each plot, constant (“ah” vocalization; left) and intermittent (two hand claps; right) time waveforms are displayed.

Fig 2. Mean ratings of valence of sounds when they were rms-scaled as a function of ratings of valence when tokens were peak-scaled.

Each dot represents an individual sound. Shaded squares indicate ratings that did not change nominal category between scaling approaches. Yellow dots indicate sounds that were nominally ‘unpleasant’ with the peak-scaling approach but were nominally ‘neutral’ with the rms-scaling approach. Blues dots indicate sounds that were nominally ‘neutral’ with the peak-scaling approach but were ‘pleasant’ with the rms-scaling approach. Red dots indicate tokens that were nominally ‘pleasant’ with the peak-scaling approach but were nominally ‘neutral’ with the rms-scaling approach.

Fig 3. Mean ratings of valence as a function of rms level (among peak-scaled sounds; left panel) and as a function of peak level (among rms-scaled sounds; right panel).

The relationship between ratings of valence and rms level is statistically significant, whereas the relationship between ratings of valence and peak level is not statistically significant.

Fig 4. Mean rating of arousal of tokens when they were rms-scaled compared to when they were peak-scaled.

Each dot represents an individual sound. Shaded regions indicate ratings that did not change nominal category between scaling approaches. Yellow dots indicate sounds that were nominally ‘neutral with the peak-scaling approach but were nominally ‘calming’ with the rms-scaling approach. Blues dots indicate sounds that were nominally ‘exciting’ with the peak-scaling approach but were ‘neutral’ with the rms-scaling approach.

Fig 5. Mean ratings of arousal as a function of rms level (among peak-scaled sounds; left panel) and as a function of peak level (among rms-scaled sounds; right panel).

The relationship between ratings of arousal and rms level was statistically significant, whereas the relationship between ratings of arousal and peak level was not statistically significant.