Conference talk clips, from the recording
A talk is given once and then lives as a forty-minute recording nobody watches to the end. The parts that made it worth giving are three or four minutes of it, spread across the whole.
A talk is structured, and the structure scores
A good talk has deliberate peaks: the line the speaker slowed down for, the demo, the point where the room laughed. Those are audible events as well as textual ones, and the audio-energy pass registers them without needing to understand them - applause and laughter are, to a scorer, unusually loud moments that were not the speaker's voice.
Combined with the transcript, that is a strong signal. It is the closest the pipeline gets to being told which moments landed by the people who were there.
Slides stay readable, within reason
Because the 9:16 output keeps the whole frame instead of cropping to the speaker, a talk recorded as a picture-in-picture or a switched feed keeps its slides. A slide with a sentence on it reads fine at that size. A slide with a code listing on it does not, and no reframing strategy would have saved it.
Publish the clips, keep the talk
Clips are ordinary MP4 files with no restrictions attached, and on a paid plan no watermark. The recording stays where it is - this produces a set of short videos from it, it does not replace or re-host the original.
How it works, step by step
- 01Upload or import the recordingUpload the recording of the talk. A static wide shot of the stage works as well as a switched multi-camera feed.
- 02Peaks are foundScene cuts, audio energy, motion and the transcript are scored together - laughter and applause register as energy alongside what was being said.
- 03Clips are assembledEach clip grows around a peak and snaps to the nearest natural pause, then gets burned-in captions from the real transcript.
- 04Review and downloadTrim anything that needs it, then download individually or as a ZIP.
Questions
Does audience laughter actually help it find moments?
It contributes. The audio-energy pass does not know what laughter is - it registers a loud moment that is not the speaker's steady voice, which in a talk recording is usually laughter or applause, and usually follows something worth clipping.
What if the recording is a static wide shot of the stage?
That works. There is little motion signal, so scoring leans on speech and audio, and the reframe preserves the whole stage rather than cropping to a figure in the middle of it.
Can I do a whole conference at once?
One video at a time. How many you can run in a month is set by your plan.