Student-teacher RCT: adding 360° VR to voice training was engaging but no better short-term than imagining the classroom

Background

Voice disorders are common in teachers, with prevalence estimates ranging from roughly 20% to 80%, and the warning signs appear early: a meaningful share of student teachers already report recurring voice symptoms before they begin their careers. Teaching places sustained, high-intensity demands on the voice in acoustically difficult rooms, and the consequences of vocal difficulty reach well beyond the voice itself, affecting absenteeism, professional identity, well-being, and even pupils’ ability to follow what is being said.

Both indirect (education-focused) and direct (exercise-focused) voice training programs can improve vocal symptoms and awareness, but two problems persist: effects are often modest in healthy trainees, and - most importantly for this study - carrying newly learned vocal techniques out of the training room and into the real classroom is hard. Immersive virtual reality has been proposed as a way to close that gap by letting people practice inside realistic, controllable simulations of the settings where the skills actually need to be used.

This study builds directly on earlier work by Remacle and colleagues, who created a computer-generated virtual classroom for student-teacher voice training. That earlier work compared an indirect program against a direct, VR-supported program, which made it impossible to isolate the effect of VR itself from the effect of the training content. Bostyn and colleagues set out to fix that confound: they held the training content constant across both groups and varied only whether the exercises were performed inside an immersive 360° school environment or while simply imagining one.

What the researchers did

This was a randomized controlled trial with a pretest-posttest control group design, run by a Ghent University / KU Leuven team in Belgium. Sixty-three student teachers enrolled in a Flemish primary- or secondary-teaching program took part (all Dutch speakers; 49 cisgender women and 14 cisgender men; mean age 28.9 years, SD 10.8). Participants were vocally healthy: anyone with self-reported hearing problems, a diagnosed organic voice pathology, recent regular smoking, or prior intensive voice training was excluded. They were organized into small training groups of four and assigned by blocked cluster randomization (with sex at birth as the blocking factor) to either the VR group (n=30) or the control group (n=33), and were blinded to the existence of the other condition.

Both groups received the same three 90-minute group voice-training sessions over three consecutive weeks, delivered by trained speech-language pathology master’s students who were themselves blinded to allocation. The sessions covered vocal hygiene plus exercises for posture, breathing, resonance (including semi-occluded vocal tract exercises), vocal warm-up, and voice projection with articulation. The single experimental difference was the delivery of the exercises:

• VR group performed the exercises while wearing a Meta Quest 3 headset, immersed in 360° videos of a classroom, cafeteria, and playground that the authors had recorded in a real Flemish primary school.
• Control group performed the identical exercises while being asked to mentally visualize the same school settings.

Background noise (for example, around 75 dB on the playground and 40 dB in a quiet classroom) was matched across the two groups, so the only thing that genuinely differed was the visual immersion provided by VR.

Before the first session and after the third, each participant was assessed individually on:

1. Self-rated vocal effort on the Borg CR10 scale, after reading a long passage under 65 dB of white noise.
2. Researcher-rated posture, muscle tension, and voice projection on a 0-100 visual-analogue scale, plus the six CAPE-V perceptual parameters (severity, roughness, breathiness, strain, pitch, intensity), scored from audio and video of a one-minute spontaneous “explain a topic to the classroom” task under white noise.
3. Acoustic measures - fundamental frequency, fundamental frequency range, and intensity (loudness) range - analyzed in Praat from sustained vowels and a read passage.
4. A training-experience questionnaire (general questions for everyone, plus VR-realism questions for the VR group and imagination-difficulty questions for the control group).

Data were analyzed with linear mixed-effects models with Cohen’s d effect sizes. The authors deliberately applied no correction for multiple comparisons and frame the trial as exploratory; rater reliability was checked with intraclass correlation coefficients.

What they found

Both groups improved on the two outcomes most sensitive to short-term training, with no VR advantage. Self-rated vocal effort fell significantly in both groups (VR -1.05, control -0.83 Borg points; medium-to-large effects), and both significantly widened their loudness range (VR +3.99 dB, control +3.35 dB; moderate effects). There were no significant between-group differences and no interactions on either measure - the two delivery methods worked about equally well.

Where the groups differed, the control group did slightly better - but with caveats. The control group showed significant improvements in researcher-rated posture and muscle tension, while the VR group did not. The authors are careful here: the control group started with more visible tension than the VR group, the two groups ended up similar after training, and several VR participants reported feeling overwhelmed or distracted by the immersive environment and noted that the headset made it harder to maintain good posture. So the “control did better” pattern is plausibly a mix of baseline imbalance, a headset-on-posture artifact, and the limited reliability of these perceptual ratings, rather than a true disadvantage of VR for voice technique.

Most other measures did not move. Voice projection did not change in either group (scores were already good at baseline), and the CAPE-V perceptual parameters were largely flat, with only “strain” improving weakly across the whole sample. Fundamental frequency and frequency range were unchanged - unsurprising, since the exercises targeted breath support and projection rather than pitch, and the participants were vocally healthy with little room to improve. For nearly all of the non-significant outcomes, the confidence intervals were wide enough that the study simply could not rule out a real effect; it was underpowered.

VR’s clearest win was experiential. Most VR participants rated the classroom (76.7%) and cafeteria (80%) as realistic and the environments as pleasant to practice in (70%); the playground fared worse (53.3%), blamed on the 360° video’s sparse, slow-reacting virtual pupils. By contrast, a large minority of the control group could not vividly imagine the school settings at all (around 42-45% for each), and those who had never taught struggled the most. Every participant rated the training as useful, two-thirds found the exercises easy, and self-reported home practice did not differ between groups.

Why this matters

This is one of the first trials to isolate the effect of visual VR immersion in voice training by holding the training content identical across groups - the confound that limited earlier student-teacher VR work. Its honest headline is a null: over three sessions, immersive 360° VR did not improve short-term voice outcomes beyond the same exercises done while imagining the classroom, and may even have nudged posture and tension the wrong way through headset interference.

The more interesting signal is about engagement and access. Asking a trainee who has never stood in front of a class to “imagine the classroom” frequently failed, whereas the VR scenes were rated realistic and pleasant. If a vivid, present-feeling context helps trainees stay motivated and practice consistently - the mechanism the authors propose - that could matter for a preventive program whose whole point is building durable habits before voice problems start. But the authors are explicit that this is a hypothesis: motivation was not a controlled outcome here, the program was short, the participants were vocally healthy, and any transfer benefit would need longer trials with real classroom follow-up to demonstrate.

For Therapy withVR specifically: this study did not use, evaluate, or compare against Therapy withVR. The system tested was custom 360° video recorded by the research team and played on a Meta Quest 3, used to coach vocal technique for voice-disorder prevention - a different domain from Therapy withVR’s focus on rehearsing speaking situations and managing speaking-related anxiety. The Bostyn paper is included in the Evidence Hub because it adds to the broader evidence base on immersive VR in speech-language pathology, not because it relates to Therapy withVR.

Limitations

The authors flag the following in their discussion:

• Limited inter-rater reliability. Agreement between the three researcher-raters was sometimes poor (ICC as low as 0.339), and they did not calibrate their ratings together beforehand. This weakens confidence in the perceptual outcomes (CAPE-V, posture, tension, projection), which the authors say should be interpreted with caution.
• Underpowered for several measures. With 63 participants split across two groups, confidence intervals for the non-significant outcomes were wide; the study could not rule out real effects it lacked the power to detect.
• No correction for multiple comparisons. With many outcomes tested and no correction applied, some statistically significant findings may be false positives. The authors explicitly call their results hypothesis-generating rather than confirmatory.
• Vocally healthy, short, preventive program. Participants had no voice complaints and trained for only three sessions, leaving little room for measurable acoustic change; the program aimed to build habits, not to remediate disorder.
• CAPE-V applied outside its clinical standard. Perceptual ratings were taken from speech produced under 65-dB white noise, which can induce Lombard speech, so the CAPE-V scores reflect relative within-study differences, not valid clinical diagnoses.
• Baseline imbalance and a possible headset effect. The control group began with more visible tension than the VR group, complicating the one borderline interaction, and the VR headset itself may have affected participants’ posture and body awareness.
• Limited auditory realism and small-group format. Participants’ own voices were not integrated with the environmental sound and room reverberation was not simulated, reducing ecological validity; training in groups of four allowed peer feedback but also added variability.
• Mostly women. About three-quarters of participants were women (who are generally more susceptible to voice disorders), which could bias results, though this also reflects the real teaching population.