OpenTranscribe — Speaker Clustering & Naming

OpenTranscribe — Speaker Clustering & Naming A workflow diagram generated by Archify. 01 / Embedding 02 / Matching 03 / Clustering 04 / Human Review Automatic Human-in-the-loop Voiceprint · new speaker embedding · Embedding › Automatic Voiceprint new speaker embedding Match · cosine vs known profiles · Matching › Automatic Match cosine vs known profiles Auto-assign · above threshold · Matching › Automatic Auto-assign above threshold Cluster · no match — group across files · Clustering › Automatic Cluster no match — group across files Profile · named, reusable SpeakerProfile · Clustering › Human-in-the-loop Profile named, reusable SpeakerProfile Review · merge / split / verify · Human Review › Human-in-the-loop Review merge / split / verify match found no match feeds future matches promote Legend Agent logic Context / trace External system

Threshold gates automation

  • • Confident matches auto-assign to an existing profile — no human needed
  • • Everything below threshold becomes an unverified speaker, never guessed

Clustering is cross-file

  • • find_or_create_cluster groups similar unverified speakers across the whole library, not per-file
  • • batch_recluster can re-run this after embeddings change (e.g. a model upgrade)

Promotion is the only way to a name

  • • merge_clusters / split_cluster let a reviewer correct grouping mistakes first
  • • promote_cluster_to_profile is the one step that creates a durable, reusable identity