Podcast clips, cut from the episode
A two-hour episode contains maybe six minutes anyone will share. Finding them means scrubbing the whole thing again, which is why most episodes get one clip cut in a hurry or none at all. Glyphcast reads the episode and proposes the moments.
Why podcasts are the easiest case
Everything the scorer is good at, a podcast has. There is continuous speech, so the transcript is dense and word-level timings are accurate. There are natural pauses between thoughts, which is where a clip should start and stop. And the interesting moments are audible - a laugh, an interruption, someone raising their voice - which the audio-energy pass picks up independently of what was said.
What it is not good at is a video with no speech in it. A podcast is the opposite of that problem.
Clips start and end where a sentence does
The failure that makes an automatically cut clip unusable is not a bad moment - it is a good moment that starts three words late and ends mid-word. Glyphcast grows a clip outward from the scored peak and then snaps each edge to a pause in the transcript, so the clip opens on the start of a thought and closes after it lands.
You can still move either edge. Trimming a clip re-runs it through the same pipeline rather than cropping the finished file, so the captions re-time to the new cut instead of drifting out of sync with it.
Video podcasts and audio-only episodes
A video podcast is the straightforward case: the source has a picture, and the 9:16 output keeps the whole frame with a blurred copy of itself filling the bars above and below. Nothing is cropped, so a two-person shot stays a two-person shot.
An audio-only episode has no picture to work with. Glyphcast is built for video sources, and an audio file with a static cover image will produce correctly cut, correctly captioned clips of a still image - which is a real format on some platforms, but it is worth knowing that is what you will get rather than discovering it afterwards.
How it works, step by step
- 01Upload the episodeUpload the episode. A video podcast is the straightforward case; an audio-only file with a static cover works, with the caveat below.
- 02Let it read the episodeSpeech is transcribed with word-level timings, and every moment is scored on what was said, how loud it got, and how much movement there was.
- 03Review the proposed clipsEach clip arrives cut to a natural pause, reframed to 9:16 and captioned. Rate them, trim the edges, or delete the ones that missed.
- 04DownloadDownload clips individually or take the whole set as a ZIP. They are ordinary MP4 files.
Questions
How long can a podcast episode be?
The ceiling depends on your plan, and it is listed on the pricing page. Longer episodes take longer to process - the transcription pass is the slow part - but length does not change how the moments are chosen.
Will it find moments in a two-person conversation?
Yes, and conversation is the material it reads best: interruptions, laughter and changes in pace all register as signal. What it does not do is crop to whoever is speaking - the 9:16 output keeps the whole frame rather than tracking a face, so both people stay in shot throughout.
Do the clips have captions?
Yes, burned into the picture and generated from the real transcript rather than from a second pass over the clip, so the words match what is actually said and land on the right frame.