← the bench

@snowai-app/transcribe

Transcription

Every word keeps its time, so a transcript can be clicked back into the audio — and the same component would serve an agent dictating a client note in GetCovered and a Docubase user dictating a document. Wanted by both; lifted for neither yet.

Running on fixtures — the recording ships inside the package; nothing is uploaded and no transcription service is called.

A synthetic ninety-second intake call: two tones, one per speaker, over exactly the transcript’s timings. Not a recording of anyone.

Press Transcribe. The fixture provider holds the job in processing for a few seconds — a real Amazon Transcribe job takes minutes, and a page has to be able to sit in that state — then returns the transcript with speaker turns and a time on every word.