scribe

This reference skill covers Zoom AI Services Scribe, a transcription API for turning uploaded or stored media into text. It helps developers design transcript pipelines around synchronous single-file (fast mode) transcription, asynchronous batch jobs, webhook-driven status updates, and Build-platform JWT authentication, and it clarifies where Scribe fits versus real-time media streaming.

The skill documents the core workflow: generate an HS256 Build-platform JWT, choose fast mode for one short file or batch mode for stored archives and large sets, submit the request, poll job/file status or receive webhooks for batch jobs, then persist and post-process the transcript JSON. It maps the endpoint surface (transcribe, jobs create/list/inspect/cancel, per-file results) and includes practical operational guidance: real deployed-sample timing observations, a note that a frontend 504 with a backend 200 is a browser/edge timeout race rather than a transcription failure, a recommendation to wrap fast mode in async request/polling for hosted UIs, and a browser-microphone pseudo-streaming pattern (short chunked uploads) explicitly framed as a fallback demo rather than a substitute for real-time streaming. It also documents credential-naming drift across Zoom docs and advises verifying labels in the portal. Its auth guidance is defensive — keep JWTs short-lived (one hour or less) and sign server-side.

It targets backend and application developers building on-demand clip transcription, batch transcription of stored archives, webhook-driven ETL pipelines, and compliance/QA transcript workflows. The content is documentation and code-pattern reference grounded in the official Zoom AI Services docs and the scribe-quickstart sample.