Breaking down a video

If the video already publishes captions, Novus Learn uses them: it is fast, accurate and needs no model. If it does not, you can transcribe an audio or video file you already have using a speech model that runs inside this browser, in a same-origin worker. The file is never uploaded.

• A one-time model download the first time you transcribe. It is large, and it is cached afterwards.
• Real processing time on your own hardware, which varies enormously between devices.
• Nothing else. There is no per-minute charge, no key, and no upload.

Once there is a transcript, the pipeline is the one every other source uses. Claims are extracted from the transcript's own sentences, and each one carries its timestamp rather than a line number, so inspecting the evidence takes you to the moment in the recording where it was said.

Automatic transcription makes mistakes, especially with names, technical terms, overlapping speech and strong background noise. Because every claim links back to its timestamp, a suspicious sentence can be checked against the audio in a few seconds. Treat the transcript as a working copy, not a record.