Hearing concussion before it's too late
A prospective study with a pharmaceutical sponsor, applying Sona-2 to real-world TBI detection in contact sport athletes. What started as video recordings from a rugby pitch in Australia is now a formal 18-week longitudinal study.
- Modality
- Voice / acoustic AI
- Condition
- TBI / concussion
- Phase
- Prospective study · active
- Location
- Australia · elite contact sport
How this partnership came together
Concussion leaves a fingerprint in the voice. Disrupted phonation, altered prosody, spectral texture changes — signals too subtle for a sideline assessment but within reach of an acoustic model trained on the right data.
The program started with a question a clinical collaborator brought to Amplifier: can voice detect the residue of a brain injury that no imaging can see? The neuromechanics are real — concussion disrupts phonation at the brainstem level, altering prosody, timing and spectral texture in ways humans cannot perceive.
The clinical team connected Amplifier with a trial infrastructure partner in Australia, purpose-built for contact sport research with direct access to elite athletes and established biomarker protocols. The trial partner had the athletes, the longitudinal blood panel design and the methodology. Amplifier had Sona-2.
The brain injury shows up in the voice before it shows up in the clinic. We just needed a model that could read it.
The two samples that started everything
Before committing to a full prospective design, the trial team sent video recordings of athletes confirmed concussed the prior season, recorded weeks post-injury. The Amplifier team stripped the audio and ran raw inference through Sona v2.1.0.
90.4% TBI positive
Signal persisted weeks post-injury with no baseline comparison and no preprocessing adjustments. Direction confirmed.
78.4% TBI positive
Lower confidence, consistent with post-acute attenuation and individual variation — still above threshold.
What that run was — and wasn't
A directional feasibility test
Raw inference on post-acute voice samples: no preprocessing adjustments, no baseline calibration, no acute reference point.
Not a validation study
Samples were captured weeks post-injury with no pre-injury baseline. The phase was designed to establish signal direction, not sensitivity or specificity.
Why it still mattered
Every sample returned TBI-positive above threshold. In the wrong direction that result kills the hypothesis. It pointed right — enough to justify the cost of a real study.
The real study. Now underway.
Longitudinal. Acute-phase capture. Multi-modal biomarker protocol. This is where the validation data gets built.
Pre-study
Protocol finalization, voice modality integration, athlete enrollment, sponsor infrastructure confirmed.
Acute phase
A confirmed concussion triggers immediate voice capture alongside blood biomarker draws, establishing the first longitudinal comparison points.
Sub-acute phase
Continued voice and blood biomarker collection with SCAT6 symptom evaluations running in parallel, tracking the recovery trajectory.
End of study
Full dataset assembled. Voice biomarker performance analyzed against blood biomarkers and SCAT6, and commercial deployment readiness assessed.
Protocol at a glance
- Population
- Confirmed concussed elite contact sport athletes
- Duration
- 18 weeks
- Comparator
- SCAT6
- Blood biomarkers
- Multiple draws, acute and sub-acute
- Voice modality
- Additional modality alongside the blood panel
- IRB
- Not required for this phase
- Location
- Australia · sponsor infrastructure
- Model
- Sona v2.1.0
Global licensing, not acquisition
Every professional contact sport franchise — rugby, football, soccer, hockey, MMA — carries the same concussion liability. The motion is an annual per-organization license: cloud-based, hardware-agnostic, integrated with EHR and team medical staff workflows, de-identified and federated so no raw audio leaves the organization. Non-exclusive by default, with per-sport and per-league volume pricing.
Where this goes from here
Full dataset assembled
Eighteen weeks of multi-modal collection across the acute and sub-acute recovery window delivers the first longitudinal voice biomarker dataset in confirmed concussion, with SCAT6 and blood panel comparators built in.
Sona-2 TBI model locked
Validation analysis — sensitivity, specificity and confidence calibration against comparator modalities — refines and locks the TBI phenotype model for deployment.
Global licensing opens
A validated model and a clean clinical evidence package open the licensing program to sport organizations globally, starting with existing engaged partners and expanding by sport and league.
If your athletes are at risk of concussion, we want to talk
Sona-2 is built for clinical deployment. Team medical staff, league safety programs and sport research groups can run it against their own protocol and comparators.
Pilot samples were collected weeks post-injury with no acute baseline and reflect directional findings only. Sensitivity and specificity are established by the prospective study.