QC-Talk — dev/test UI

Manual browser client for the streaming transcription + extraction endpoints — "mute mode": this page never talks back, it only listens to whoever's actually speaking (e.g. a real doctor and patient) and, if you give it a Doctor ID it knows, labels who's who. Not a doctor-facing product screen — this exists to test the service, nothing more. Want the AI to actively ask the patient questions instead? Virtual doctor chat is here or talk to it live here.

Prototype only. No auth, no session ownership, nothing here is saved anywhere. Extracted vitals/symptoms are suggestions to review, not a chart entry — see the README before trusting a value you see below.

Doctor voice enrollment (learn once, reused every future session)

Have the doctor speak naturally for 15–30s (their name, a sentence or two — content doesn't matter, just clear speech). This never leaves your machine — it's a local voice-embedding model (app/diarization.py), not a call to OpenAI. Enroll once per Doctor ID; every later session that uses the same Doctor ID (here or via the API) automatically labels segments "doctor"/"patient" by comparing against this — no re-enrollment needed.

not started — enter a Doctor ID above first

Live session

If this Doctor ID has an enrolled voice (see "Doctor voice enrollment" below), the transcript's lines get labeled doctor/patient automatically — leave it blank to transcribe without speaker labels, same as before.
idle
Laptops (especially ones with array mics/OEM audio software) often expose more than one input device, or ship with aggressive noise suppression that can crush voice input before it reaches the browser at all - if the level meter below stays near-silent even close to the mic in a quiet room, try a different device here before assuming anything else is wrong.
mic level: —

Transcript

(nothing yet — click "Start talking" and allow microphone access)

Consultation note ("mute mode" summary of what was actually said)

Unlike the transcript above, this step DOES send the transcript text to OpenAI (app/ambient_scribe.py) to write a structured summary — same third-party-API caveat as the virtual-doctor pages. This AI never asks anything and never spoke to anyone; it only reads a finished transcript once.

finish a session with some speech in it first

Extraction result

(will populate automatically when the session ends)

Test /extract directly (no mic needed)