Verify a realtime speech-to-speech model by opening a session for it (voice realism VR1, D9)

Shipped

October 26, 2026 at 8:28 PM UTC

Author

Kamo

Commit

6d1b8ed

ProbeRunner runs the new REALTIME_SESSION probe (kamo-shared-library 1438893a) for REALTIME_AUDIO: it mints a 10-minute client secret for the model on the and passes when the vendor opens it; the secret is never used or kept. Until now an org's realtime model could only be attested, which org providers cannot be, so an SI never got a speech-to-speech ticket.