Find the [minute] worth sharing.
Momentize listens to an episode and hands back ranked clips with clean in and out points, plus chapters, speakers, topics, and ad reads.
[Starts on a pause of 0.90 s
]Lets the laughs settle
Everything an editor looks for, with timestamps
Ranked clips
Scored for how worth posting they are, with the reasons listed: a payoff, laughter, a quotable line, and what counts against it.
Clean edges
Each clip's start and end land between words, on a pause where there is one, and let a laugh finish. Each edge says what it snapped to.
Chapters
Where the conversation changes subject, with a title and a one-line summary.
Speakers
Who's talking and when, named from how they're introduced, with each person's share of talk time.
Ad reads
Host reads and inserted spots. Clips that would run into one are cut short or dropped.
Topics, style, and quality
What the episode covers, how it's paced, and its loudness and peaks.
From feed to clips
Point it at a feed or a file
Paste an RSS feed, an audio URL, or upload a file. Every timestamp belongs to the exact file downloaded.
It listens
The waveform gives pauses, loudness, and peaks. A speaker-labelled transcript with word timings gives the words and the laughs.
It reads
A language model names speakers, finds chapters, topics, and ads, and picks candidate moments, quoting the lines it means.
It sets the edges
Code turns those lines into times, snaps each edge between words, keeps clips out of ads, and ranks them.
Tobias
Mira Holt
Show pages from your feed
From the feed and the analyzed episodes, Momentize builds a page for the show: the people with their roles, talk time, and best clips, the show's structure, an episode list, and a page per clip. Claims about people cite the transcript lines they come from, and a claim with nothing to cite is left out.
Questions
Does it post clips for me?
No. You get clips, timestamps, reasons, and previews; you choose what to publish and where.
What kind of shows work best?
Conversation and interview shows in English with a few voices. Files can be MP3, M4A, WAV, or FLAC, up to 3 hours.
What about ads inserted at download time?
Inserted ads change from one download to the next, so every timestamp is tied to the exact file that was analyzed, by its SHA-256. If a later download differs, Momentize says so instead of pointing at the wrong moment.
How accurate is it?
Not measured yet. Every output carries the model's own confidence, which isn't the same thing. Accuracy will be measured by people grading outputs on shows the system wasn't tuned on, and published with its margin of error.