Audexia turns clinical voice into a governed, FHIR‑native artifact — protocol‑based capture, on‑device feature extraction, and a durable record of derived acoustic features.
Clinicians order a task bundle. Patients complete it on a phone. Audexia extracts acoustic features on the device, destroys the raw audio, and writes a structured, longitudinal artifact back to the EHR — features only, no recordings retained.
Choose by condition — ALS, Parkinson's, frailty, cognitive screening, mood — or compose your own. Audexia auto‑selects evidence‑based tasks (sustained /a/, DDK pa‑ta‑ka, read‑aloud, picture description).
A QR code or text link opens a guided session. Acquisition‑time QC gates — silence, clipping, SNR, prompt adherence — accept or re‑prompt before the session ever leaves the device.
Acoustic features are extracted on the patient device. The raw waveform is destroyed at the edge — only the derived feature vector reaches downstream models. Audexia does not own or host inference models. A FHIR Observation is written back to the chart. No recordings retained, by design.
Audexia computes the same evidence‑based features researchers cite — grouped by speech subsystem so clinicians can interpret the score, not just the number.
Audexia represents every captured session as a FHIR Media + Observation + Provenance bundle. The canonical artifact is the feature vector — never the recording — so the science can evolve without re‑capturing patients.
MediaThe canonical feature vector for the session — task, codec, sample rate, and quality metadata. Raw audio is never persisted.ObservationPer‑feature scores, condition‑specific composites, and confidence — all coded.ProvenanceCapture device, model version, consent, and chain of custody — fully auditable.DocumentReferenceOptional indexing layer for EHRs that prefer document‑oriented retrieval.Voice is biometric data. Raw audio will never be an issue — it never leaves the patient's device. Audexia is built HIPAA‑first, with privacy and compliance as core design constraints, not afterthoughts.
Audexia is designed for regulated clinical environments. We execute BAAs with covered entities and operate with HIPAA compliance as a baseline, not an option.
Features are extracted on the patient device. The raw waveform is destroyed before anything leaves it — Audexia never receives, transmits, or stores voice recordings.
All data is encrypted in transit (TLS 1.3) and at rest (AES‑256). Feature vectors and FHIR artifacts are the only things that ever cross the wire.
Per‑task consent is captured at the device. FHIR Provenance records bind capture metadata, software version, and clinician identity — every score is traceable back to its source.
Audexia is in early access with a small group of design partners across neurology, geriatrics, behavioral health, and research. Pricing is per‑program — scoped to your task bundles, capture volume, and integration footprint.