OpenTranscribe — Cloud ASR & Voiceprinting

OpenTranscribe — Cloud ASR & Voiceprinting A workflow diagram generated by Archify. 01 / API / Web 02 / Cloud ASR Provider 03 / Speaker Identity Transcribe Voiceprint + match Upload · user's audio/video · API / Web › Transcribe Upload user's audio/video Provider Factory · pick configured provider · API / Web › Transcribe Provider Factory pick configured provider Merge · cloud transcript + speaker turns · API / Web › Voiceprint + match Merge cloud transcript + speaker turns Cloud ASR · AssemblyAI / AWS / Azure / etc. · Cloud ASR Provider › Transcribe Cloud ASR AssemblyAI / AWS / Azure / etc. Voiceprint · CPU-queue embedding extraction · Speaker Identity › Voiceprint + match · dedicated task Voiceprint CPU-queue embedding extraction dedicated task Match · same pipeline as local ASR · Speaker Identity › Voiceprint + match Match same pipeline as local ASR transcript (+ speaker labels, if any) downloaded audio still available Legend Agent logic Context / trace Cloud service External system

Voiceprinting already covers cloud ASR

  • • No native PyAnnote centroids from a cloud transcription, so extract_speaker_embeddings_task runs on the CPU queue instead
  • • Same downstream matching/clustering pipeline either way — see the Speaker Clustering diagram

Not every provider diarizes

  • • Providers that return speaker labels: merged in directly
  • • Providers that don't: local PyAnnote diarization still runs on the downloaded audio to produce turns

8 providers behind one factory

  • • AssemblyAI, AWS, Azure, Deepgram, Gladia, Google, OpenAI, Speechmatics — plus local
  • • Swapping providers doesn't change anything downstream of Merge