VR audiences raise subjective distress but not physiological arousal or stuttering frequency in adult males who stutter

Background

Studies of arousal in adults who stutter have used multiple measurement channels - subjective self-report (e.g., SUDS), autonomic-nervous-system measures such as heart rate (HR) and skin conductance level (SCL), and HPA-axis measures such as cortisol. Across the broader social-anxiety literature, these channels do not always agree: socially anxious individuals giving speeches often report increased subjective distress without corresponding increases in HR. Prior work with adults who stutter using virtual environments (Brundage et al., 2006; Brundage & Hancock, 2015; Brundage et al., 2007) had collected self-reports and behavioral measures but not physiological reactivity during the VR tasks themselves. This study set out to fill that gap by collecting subjective and objective measures simultaneously during VR speaking tasks.

What the researchers did

Ten adult males who stutter (median age 26, range 18-51) were recruited from speech and hearing clinics and local National Stuttering Association support groups in the Metropolitan Washington, DC area. Participants completed the SSI-4, OASES, S-24, FNE, and STAI-T self-rating scales before the speaking tasks. A certified speech-language pathologist verified the presence of stuttering.

Three virtual reality environments developed by Virtually Better, Inc. were used: an orientation environment (driver’s seat of a virtual parked car with ambient sounds) and two experimental environments - a Virtual Audience (~30 seated listeners, varied ethnicity, both sexes, in a medium-sized room; participant standing behind a virtual podium) and a Virtual Empty Room (the same room with empty chairs). The researcher could trigger audience reactions via keyboard hot keys (whispering, yawning, puzzled facial expressions, falling asleep, slight head-nodding); each participant experienced exactly one whispering event and one falling-asleep event at similar time points during their speech. VR equipment was a Dell Precision 390 desktop, eMagin Z800 head-mounted display, and noise-cancelling headphones.

Physiological data were recorded with a BIOPAC MP150 system running AcqKnowledge 4.1: heart rate from electrocardiogram leads on the right wrist and left ankle; skin conductance level from electrodes on the middle and index fingers of the right hand. Baseline HR and SCL were calculated from 30-second silent intervals between repeated readings of the Grandfather Passage, and percent-change-from-baseline (HR% and SCL%) was computed in 10-second increments throughout each 4-minute speech.

Each participant gave two 4-minute speeches (counterbalanced order: half audience-first, half empty-room-first) on generic topics (favorite restaurants, sports, hobbies, books) that were not known in advance and required no preparation. During each speech, participants reported their subjective distress using the Subjective Units of Distress Scale (SUDS; 0 = extremely calm, 100 = extreme amount of distress) at one-minute intervals - four SUDS ratings per speech. Stuttering frequency was coded post-hoc from audio recordings by a research assistant using CLAN/CHAT conventions; inter-judge agreement was 92.7%.

Statistical analysis used non-parametric Wilcoxon signed-ranks tests (due to non-normal distributions) with a Bonferroni-corrected significance threshold of 0.012 (0.05/4 comparisons across SUDS, HR%, SCL%, %SW).

What they found

SUDS ratings were significantly higher in the Virtual Audience condition than the Virtual Empty Room (median 30.0 vs 20.0; W = 36, p = 0.011 against the corrected 0.012 threshold; medium effect size r = 0.57). SUDS ratings did not differ significantly across the four minute-by-minute time points within each speech, so the authors used the post-1-minute SUDS for the between-condition analyzes.

Neither physiological measure reached significance. HR% (median -1.73 Audience vs -0.46 Empty Room; W = 39, p = 0.24) showed no difference. SCL% (median 11.18 Audience vs 6.95 Empty Room; W = 6, raw p = 0.02) was in the expected direction (higher with audience) but did not reach the Bonferroni-corrected threshold of 0.012; the authors note “considerable variability in the SCL measure across participants.”

Stuttering frequency (%SW) did not differ between conditions (median 4.35 Audience vs 5.03 Empty Room; W = 17, p = 0.29).

Within each condition, the authors examined whether subjective and objective measures tracked one another. They did not: in the Audience condition, SUDS was not significantly correlated with HR% (rho = 0.32, p = 0.37) or SCL% (rho = 0.30, p = 0.40); in the Empty Room condition, the corresponding correlations were also non-significant (HR% rho = 0.17 p = 0.65; SCL% rho = 0.49 p = 0.15). Stuttering frequency was also not significantly correlated with SUDS, HR%, or SCL% in either condition.

The headline result is therefore a clear subjective-objective dissociation: the virtual audience reliably elevated self-reported distress (with a medium effect size and at the corrected significance threshold) but produced no corresponding change in autonomic arousal or stuttering frequency.

Why this matters

The dissociation argues against relying solely on physiological biofeedback or %SS counts to gauge clinical response to a VR audience exposure: a clinician using only objective metrics could conclude the audience condition had no effect, when in fact the client experienced substantially more distress. The authors propose that VR audiences may be particularly well suited to interventions that target the subjective, avoidance, and acceptance dimensions of stuttering (e.g., Van Riper or Sheehan-style avoidance reduction), where the SUDS-channel responsiveness shown here is the relevant outcome.

The paper also adds a methodological data point to a broader pattern in the social-anxiety literature: subjective and objective channels often diverge during speaking tasks, and one’s cognitions about a speaking situation may not match the body’s response. Both should be measured to characterize a client’s full reactivity profile.

Limitations

The authors explicitly flag the following in their discussion:

• Small, all-male sample. Only 10 participants; only men. The authors state this limits the generalizability of findings.
• High variability in SCL across participants. Similar to other studies (Weber & Smith, 1990). The SCL% comparison was in the expected direction but did not reach the corrected significance threshold; with a larger sample it might.
• Speech topics varied across participants. Although topics were all generic, the variability may have contributed to between-participant differences. The authors suggest future studies limit topics or use re-tell tasks.
• Sample had high STAI-T but relatively low FNE. The authors note that future studies should specifically recruit PWS with high FNE to empirically test the value of VREs in that subgroup.
• The two VR environments were intentionally minimally differentiated (same room, audience present vs absent) to provide a controlled comparison. This is a methodological strength for isolating the audience-presence variable but it also constrains how strongly the authors can claim that “VR can manipulate distress in PWS” - the effect shown here is specifically audience-vs-no-audience in the same physical setting.
• No meaningful-change benchmark for SUDS in PWS. The authors note that future research is needed to establish what constitutes a clinically meaningful change in SUDS for adults who stutter.