A second data stream between formal assessments.

Voice is the cheapest longitudinal instrument available. A participant can produce a usable sample from a phone in under a minute, at whatever cadence your protocol needs, without travelling anywhere. What has been missing is a model trained on enough labelled clinical audio for the resulting signal to be worth analysing.

Two paths. The existing signal library, applied to your cohort at whatever cadence you set — useful where an established sign maps onto your question.

Or custom inference, where we fine-tune the base model on your voice data toward a signal you define. New endpoints, new populations, indications with no existing acoustic literature. You get a dedicated endpoint. Model weights stay with us.

Decentralised collection, site-based collection, or both. Sample capture works from a standard phone.

Grant-funded and government research programmes are served through the same path. We have existing familiarity with the contracting structures.

A new custom signal is not a validated endpoint on day one. It is a hypothesis with a model behind it, and establishing validity is the work your programme exists to do. We are explicit about this at scoping because the alternative is a difficult conversation later.

We do not run your analysis, own your data, or take authorship positions by default.