Voice scoring

The athlete speaks. The phone scores.

Voice recognition is what makes self-administered cognitive testing actually work. HQ uses on-device speech recognition to capture and score recall, digit span, and months-backward responses in real time.

Book a demo How it works

Why voice matters

Traditional paper-based sport concussion batteries require a proctor to read a word list aloud, listen to the athlete’s response, and tally correct words on a paper form. It’s slow, prone to bias, and the single biggest bottleneck in running baselines at scale. You need one trained clinician for every athlete.

Voice scoring removes the proctor. An athlete holds the phone, hears the word list through the speaker (or earbuds), and speaks the words back. On-device speech recognition matches each spoken word against the expected list and marks hits, misses, and substitutions.

What we built

On-device speech recognition — low latency, no cloud round-trip
English and Spanish voice prompts and scoring today
Replayable audio so athletes can hear instructions again
Tap-to-confirm fallback when noise or accent produces uncertainty
Per-word confidence scores audit-logged alongside the final answer

Accessibility

Every voice-prompted step has on-screen text, adjustable reading speed, and replayable audio. Athletes with learning differences, hearing differences, or speech differences can complete the battery with equivalent accuracy.

Privacy

Voice data is processed on-device by default. Raw audio is not stored unless the athletic trainer explicitly opts in for quality assurance, and in that case it’s encrypted at rest and deleted after a configurable retention window.

Frequently asked questions

Does it work with accents?

Yes. The underlying speech models handle common US English and Latin American Spanish accents reliably. If confidence is low on any word, the athlete is asked to confirm the response on-screen.

What about athletes who can't speak clearly right now?

The tap-to-confirm fallback lets athletes complete every module without voice — they hear the prompt, tap each recalled word from a list. Scoring is equivalent.

Is audio sent to the cloud?

By default, no. Recognition runs on-device. Raw audio is only stored if the clinic explicitly enables it for QA, and in that case it's encrypted and automatically deleted.

See it in action.

Book a demo See pricing