Skip to main content
Every video’s full audio track is transcribed by Whisper (medium) during processing, producing timestamped speech segments stored in Postgres. This happens automatically on every upload — there’s no separate transcription step to trigger.

​
Retrieving transcript content

There’s no standalone “get the full transcript” endpoint. Transcript text comes back as the text_content field on transcript-type segments in search results — you get the parts that matched your query, not a sequential export of everything said. If you need the complete transcript of a video regardless of search relevance, run a broad transcript-only query (modes=transcript) and page through the results.

​
Language

Transcription itself (Whisper) supports around 99 languages. Full-text search over the transcript, however, is currently configured for English (websearch_to_tsquery with an English text-search config) — non-English transcripts are stored correctly, but text search over them may return incomplete or no matches. If your content is primarily non-English, rely more on visual search (language-independent) until broader language support ships.

​
What’s not included

  • No speaker diarization — segments aren’t attributed to individual speakers
  • No profanity filtering or redaction — transcripts are stored as Whisper produces them
  • No manual transcript editing via the API — if a transcription is wrong, there’s currently no correction mechanism
See Transcript Search for how to query transcript content, and Combined Queries for how transcript matches merge with visual and scene matches.