How to Record and Transcribe Meetings with Deepgram, AssemblyAI, Speechmatics or Soniox

LLazare Rossillon

Use your own speech-to-text provider on Zoom, Google Meet and Microsoft Teams meetings: one bot API records the call, your provider transcribes it. Setup, code and pricing for every supported provider.

How to Record and Transcribe Meetings with Deepgram, AssemblyAI, Speechmatics or Soniox
August 13, 2026

Speech-to-text providers are excellent at turning audio into text — and completely unable to get the audio out of a Zoom, Google Meet or Microsoft Teams call for you. If you've tuned a custom vocabulary on Deepgram or built your pipeline around AssemblyAI's audio intelligence, the missing piece is meeting capture, not transcription.

Meeting BaaS closes that gap: a bot records the meeting, and your provider transcribes it. One field in the request switches providers.

Supported providers: Gladia (built in, no account needed), Deepgram, AssemblyAI, Speechmatics and Soniox — plus ElevenLabs for real-time streaming transcription.

How it works

  1. Your backend creates a bot with a meeting URL and a transcription_config
  2. The bot joins the call like a participant and records it
  3. The audio is transcribed by the provider you selected — with your API key if you bring one
  4. Your webhook receives the transcript in the provider's native format, plus the recording, participant list and speaker timeline

Pick your provider in one field

curl -X POST "https://api.meetingbaas.com/v2/bots" \
     -H "Content-Type: application/json" \
     -H "x-meeting-baas-api-key: $API_KEY" \
     -d '{
           "meeting_url": "https://meet.google.com/abc-defg-hij",
           "bot_name": "Transcription Bot",
           "transcription_config": {
             "provider": "assemblyai",
             "api_key": "'"$ASSEMBLYAI_API_KEY"'"
           }
         }'

Swap "assemblyai" for "deepgram", "speechmatics" or "soniox" — nothing else changes. Provider-specific options (language hints, custom vocabulary, regional endpoints) pass through custom_params, and a region field covers providers with regional deployments.

Why bring your own key?

  • Keep your provider relationship: negotiated rates, custom models and vocabulary tuning carry over untouched.
  • Native output format: the raw transcription artifact mirrors your provider's response structure, so parsers and downstream tooling built for that provider keep working.
  • Lower per-hour cost: BYOK transcription is 0.05 tokens per recorded hour (Pro plans and above), versus 0.25 tokens per hour for the built-in option.

Or skip the account entirely

The default is Gladia, built in with no provider account: 100+ languages, automatic language detection and speaker diarization at 0.25 tokens per hour. It's the fastest path from zero to transcripts.

Provider guides

Step-by-step tutorials per provider

What this costs

Recording is 1 token per hour ($0.35–0.50 depending on token pack size), speaker diarization included. Add 0.25 tokens per hour for built-in Gladia transcription, or 0.05 tokens per hour with your own key. Time the bot spends waiting to be admitted is billed as recording time.

Next steps

Similar blogstutorial