Semantic search across your video transcripts
Ask a question in your own words. VTI matches it against the timestamped transcripts you have produced and points to the moment that answers it.
Personal testing · closed beta
How the search works
Search is semantic. Each transcript segment is embedded, so a query can match passages that describe the same idea in different words.
Each result carries its media title, its language, its start and end time and a link that opens the source at that moment.
When VTI does not answer
If no passage is relevant enough, VTI says so instead of guessing. An answer is returned only with the passages it was drawn from as citations.
Rights and limitations
Search covers transcripts of media you own or are authorised to use.
Retrieval quality depends on transcript quality. Check the cited passage before you rely on an answer.
Questions
Is the search keyword matching?
No. Search uses embeddings of the transcript segments, so it can find a passage that expresses the same idea in other words.
Does VTI always give an answer?
No. When the transcript does not support an answer, VTI reports that it abstains rather than inventing one.