Why one microphone is not always enough
On-device voice identity lets a robot decide locally who is speaking and whether the voice is live. For most commands that is the right level of assurance. But some actions — unlocking a door, authorising a payment, overriding a safety limit — deserve more than one opinion, especially when the machine is in a shared or public space.
The question is how to add assurance without adding friction to every interaction.
How trusted-device verification works
- A person speaks a command; the robot or device hears it locally.
- The machine’s own on-device voice trust scores the voice: who is speaking, is it the enrolled owner, is the speech live, replayed or synthetic.
- For a sensitive action, a trusted personal device that heard the same interaction runs its own verification of the enrolled speaker.
- The trusted device binds its result to that interaction and signs it, then hands the attestation to the machine.
- The machine combines both results and decides: allow, challenge or deny.
The five properties that matter
| Property | What it means |
|---|---|
| Independent | The robot and the trusted device evaluate the interaction separately. |
| Local | Models run on-device; raw audio doesn’t need to leave either device. |
| Cryptographic | The trusted device binds its verification to the interaction and signs the result. |
| Offline | Devices exchange the trust result locally, with no internet round trip. |
| Risk-based | Normal interactions stay frictionless; sensitive actions can require the extra factor. |
What trusted devices do — and do not — promise
A trusted device provides an independent verification signal for higher-assurance actions. It does not guarantee that a speaker is genuine; it makes an attack substantially harder, because the attacker would need to defeat two independent checks on two devices at once. That framing matters for safety cases and for honest product claims.
Getting started
Start with Vosco directly on the machine, then add trusted-device verification for the actions that need a stronger level of assurance. The architecture is shown on the Physical AI page.
Frequently asked questions
What is a trusted device in voice verification?
A trusted device is a personal device such as a phone, watch or earbuds that independently verifies the enrolled speaker during an interaction and sends the machine a signed attestation.
Does trusted-device verification need an internet connection?
No. The devices can exchange the signed trust result locally, without an internet round trip.
Is every voice command checked by the trusted device?
No. It is risk-based: everyday commands stay frictionless, and the second factor is requested only for sensitive actions.
Verify the voice before your AI acts.
Start on the Vosco API with 1,000 minutes free every month, or talk to us about on-device deployment.