captionhub

Creating and editing captions

Speaker detection

📢
Requirements: Professional or Enterprise plan
 

Speaker detection is the ability to define who is speaking at any one point in time, and to attach that information to CaptionHub's output.

 

It's currently only available when the Speechmatics transcription engine is selected. When Speechmatics is selected, then speaker detection is automatically enabled. CaptionHub can also name speakers automatically. See Automatic speaker attribution below.

 

Once speakers are detected, you can:

  1. Review and edit speaker names in the Caption Editor. See Caption Metadata section.
  1. Include speaker names when exporting. Only the DOCX (Word) transcript export reflects changes made to speakers.

Automatic speaker attribution from captions

Requirements: Professional or Enterprise plan, the Speechmatics transcription engine, and an AI connector (OpenAI or AWS Bedrock) set up for your team.

Speaker detection splits your media into speakers, but by default it labels them generically (for example, Speaker 1, Speaker 2). Automatic speaker attribution goes a step further. It reads what is said in the captions and replaces those generic labels with the speakers’ real names wherever it can.

CaptionHub looks for names at the start and end of each speaker’s turn, where introductions and handovers usually happen. For example: “Hello, my name is Will Butler-Adams…”, “Thank you, Marta, for the welcome” or “Our next speaker is…”.

Your team’s AI connector checks whether each name belongs to the current, previous or next speaker, and CaptionHub labels the speakers to match. Only these few captions are sent to the AI provider, not the full transcript.

  1. Go to Connectors > AI connectors and enable OpenAI, AWS Bedrock, or both.
  1. Choose your default AI provider in Team Settings > Default providers.
  • Names must be spoken. If no one says a speaker’s name near the start or end of a turn, that speaker keeps their generic label.
  • Roman-script languages only. Languages written in other scripts, such as Arabic, Chinese, Japanese or Cyrillic, aren’t supported.
  • Results depend on the transcription. If a name is mis-transcribed, the attributed name will be wrong too. Correct the spelling in the captions or in the speaker metadata.
  • Always review the results. The AI can occasionally assign a name to the wrong speaker, for example when several people are named in the same sentence. You can check and edit speaker names in the Caption Editor (see the Caption Metadata section).
Did this answer your question?
😞
😐
🤩