Skip to content
Studeia Docs

Live class recording and transcription (multi-provider)

How Studeia handles live class recording and transcription: role-based secure access, native-first transcription (provider VTT) with STT fallback, review and RAG ingestion.

2026-06-22 7 min
Resposta curta

Studeia handles live class recording and transcription multi-provider. Recording access is resolved by role on the server — never as a raw link, without exposing the provider key. Transcription is native-first: it uses the provider's native transcript (VTT) when available and falls back to STT (Whisper→Google) only when needed. After review (pending→review→approved→ingested), the transcript can be ingested into RAG, making the AI tutor also answer based on the spoken class.

Live classes produce two valuable assets: the recording and the transcript. Studeia handles both securely and multi-provider, and even turns the transcript into material for the AI tutor.

Quick answer

  • Recording access resolved by role on the server (admin/teacher/enrolled student)
  • Never a raw link — ephemeral token or proxy, without exposing the provider key
  • Native-first transcription: provider VTT first; STT (Whisper→Google) only if needed
  • Review flow (pending→review→approved→ingested)
  • An approved transcript can be ingested into the course RAG

Secure recording access

The recording isn't delivered as a raw URL (which would leak credentials). The platform resolves authentication behind the scenes, by provider:

ProviderHow it's delivered
BBBRedirect to the player page
ZoomRedirect with a re-minted token
TeamsProxy the bytes (Graph requires Bearer)
MeetRedirect governed by the Drive ACL

Role-based authorization (institutional admin, owning teacher, enrolled student) is checked on the server.

Native-first transcription

  1. Native first: tries the provider's transcript (Zoom/Teams/Meet as VTT), normalized to text. Free.
  2. STT fallback: only when there's no native transcript, it downloads the media (with auth resolved by the adapter) and transcribes with Whisper, falling back to Google Speech-to-Text.
  3. Limitation: default BBB may not expose downloadable media for STT — in that case, it degrades with a clear message.

From transcript to AI tutor

The transcript goes through a state flow:

pending → review → approved → ingested

Once approved, it can be ingested into the course RAG. From then on, the AI tutor also answers based on what was said in the live class — not just the written material. This broadens the tutor's coverage without manual transcription work.

Reliability

Recording webhooks are unreliable (delays, failures). So a sync cron acts as a fallback: it periodically fetches pending recordings and transcripts, ensuring nothing is lost.

FAQ

Who can access the recording? Admin, owning teacher and enrolled student — by role, on the server.

How is transcription done? Native-first (provider VTT); STT (Whisper→Google) only if needed.

Does it become tutor material? Yes, after approval, ingested into RAG.

Does it work with any provider? Yes (BBB/Zoom/Teams/Meet), with provider specifics.


See live classes and RAG ingestion.

FAQ

Who can access a live class recording?

Access is resolved by role on the server: institutional admin, the teacher who owns the class, and enrolled students (in the class, when the class is class-scoped). The recording is never delivered as a raw link — the platform resolves authentication behind the scenes (redirect with an ephemeral token or proxying the bytes), without exposing the provider key to the browser.

How does class transcription work?

It's native-first: the platform first tries the provider's native transcript (Zoom/Teams/Meet as VTT, normalized to text), which is free. Only when there's no native transcript does it fall back to STT (Whisper, with a Google Speech-to-Text fallback), downloading the media with authentication resolved by the provider adapter.

Does the transcript become study material for the AI tutor?

It can. The transcript goes through a review flow (pending→review→approved→ingested). Once approved, it can be ingested into the course RAG, making the AI tutor also answer based on what was said in the live class — not just the written material.

Does it work with any video provider?

Recording and transcription use the multi-provider adapter pattern (BBB, Zoom, Teams, Google Meet). Each provider has specifics: native transcription varies, and default BBB may not expose downloadable media for STT, in which case the platform degrades with a clear message. A sync cron acts as a fallback to webhooks.

Veja tambem

Live class recording and transcription (multi-provider)