Entrevoce - Live transcription

Entrevoce Live transcription

Idle

00:00 0 turns

Back to site

PROTOTYPE. Not authorised for real patient information. This page transcribes with the browser Web Speech API, which sends the audio to third-party servers outside Entrevoce and with no HIPAA BAA. Do not use it with PHI or on a real clinical call: demonstration and testing with made-up data only.

What that means in practice

Microphone audio leaves this machine for the browser speech recognition service. Entrevoce does not control that journey, nor its retention, nor where it lands geographically.
The transcript lives only in this tab memory. It is not written to disk, not stored in the browser, and not sent to any Entrevoce server. Closing or reloading loses it.
If you download or copy the transcript, that file becomes your responsibility. With real data it would be PHI.
For a real call the route is the Entrevoce engine with vendors under BAA, or local voice inside the hospital network.

Doctor

English · en-US

Nothing said yet.

Patient

Spanish · es-MX

Nothing said yet.

The columns are languages, not people. The console tells which language was spoken, not who spoke. If the doctor answers in Spanish, that sentence goes to the patient column. Real speaker detection (voice print, diarization) lives in the engine and needs a server; it does not exist here.
One language is heard at a time. The browser does not allow two recognition sessions. Switching language leaves a gap of a few tenths of a second with nothing transcribed, and the first sentence after each change of speaker usually lands below, unattributed. That is the price of not inventing an attribution.
One- and two-word answers are never routed automatically. "Si", "ok", numbers and proper names are undecidable between the two languages. Sentences that mix English and Spanish are not routed either: that is not a failure, it is a question with no single answer.

Title

Doctor

Patient

Unattributed 0