A five-minute interview is easy to record and hard to post. Reels, Shorts and TikTok want 15 seconds, and finding the right 15 seconds usually means watching the whole thing with a notepad, writing down timecodes, and then scrubbing back to each one. The video editor in the osFoundry desktop app can hand you a shortlist instead. It reads the interview's transcript and proposes three 15-second stretches to start from. The example below is a 5:07 NASA interview with astronaut Christina Koch, a week after her 328-day stay on the International Space Station.
Transcribe the interview
Add the interview to a project in the video editor, select the clip on the timeline, and click Transcript under the preview. The panel that opens has one button.
Click Transcribe. The editor takes only the sound from the clip, converts it to 16 kHz mono and sends that to the transcription model you set up in the app, with your own key. That is the one step that costs anything. The line under the button says so: editing, captions and clip suggestions are free.
What comes back is not just text. Every word arrives with its start and end time, and that is what the next step works from. This interview came back as 1,007 words.
Ask for suggestions
Click Suggest clips in the panel's toolbar. Three time ranges appear under the transcript, each 15 seconds long:
The editor builds a 15-second window starting at every word, counts how many words fit inside it, and keeps the three busiest windows that don't overlap. This interview averages 3.3 words a second. The three suggestions run at 4.27, 4.47 and 4.33 words a second, so they are the stretches where the most is said.
Because it is counting and nothing else, no model is involved and nothing is uploaded. The panel is also honest about what the numbers mean. Its caption reads "Densest stretches of speech — a starting point, not a verdict." Dense speech tends to be someone answering a question they care about, but the count can't tell an anecdote from a list of dates.
Jump to a suggestion and listen
Click a suggestion. The playhead and the preview jump straight to its start, so you hear it in place instead of hunting for the timecode.
Here the third suggestion, 2:31.5, is Koch explaining that she never had time to get bored on the station, with the science, the maintenance and the experiments the crew run on themselves filling her days. That is a self-contained thought and a good post. The first suggestion, at 0:45.1, catches the end of the host's question before Koch's answer, which can work too if the question gives the clip context.
Treat the 15 seconds as a rough cut. Listen through, then move the start back to where her sentence begins and the end to where it lands, so the clip opens and closes on a complete thought. A clip of 13 or 18 seconds that sounds finished beats an exact 15 that cuts a sentence in half.
Without the app
The same shortlist can be built by hand from any transcript with word timestamps. Whisper produces them, and so do most hosted speech-to-text APIs when you ask. A short script then slides a 15-second window across the words and keeps the busiest ones. Whatever produces the shortlist, the listening step stays the same.
Transcription runs on whichever provider and key you add to the app, as described on the page about bringing your own cloud API key to osFoundry. The desktop app itself is free from the osFoundry download page for macOS, Windows and Linux.
Top comments (0)