One controlled edit · September 2026
What one three-minute interview edit showed.
If you publish an interview every week, the last check matters as much as the first cut. We tried the same three edits in Descript Free and a free local workflow, then checked the exported audio and captions.
The recording was an original, fictional 191.6-second interview with synthetic voices. It included a spare lead-in, a restarted answer, an abandoned closing thought, and a one-second pause that should stay. This is one scripted exercise, not a test of a real episode or a typical user's speed.
The narrow result
Both workflows made the three intended removals. Checks of the exported files found the unwanted phrases absent, the neighboring answers present, and the planned pause retained. These were transcription and signal checks; nobody recorded a listening judgment of the joins or naturalness.
| Check | Descript Free | Free local workflow |
|---|---|---|
| Edited WAV length | 172.880 seconds | 170.520 seconds |
| Intended removals | Three supported by exported-audio checks | Three supported by exported-audio checks |
| First-pass caption fixes | Four target occurrences | Three target occurrences |
| Planned pause | One-second gap supported by waveform alignment | One-second silence confirmed in export |
The different lengths do not tell us which edit sounds better. The free local method was a technical, scripted workflow on an existing machine, not an off-the-shelf novice setup.
Captions were the clearest repair task
We checked six exact names and numbers, including “Maya Chen,” “Qadir,” “East Pier,” “Episode 17,” “3.5 seconds,” and “$12.” Descript's first-pass captions needed four corrections among those targets; the local first pass needed three. Those counts describe this recording only. They are a useful reminder to search the finished captions for the words your episode cannot afford to get wrong.
A file limitation: the four Descript corrections were entered and verified in its editor, but the final SRT/VTT download failed. Our corrected subtitle file was made locally from the raw Descript SRT. We cannot present it as a downloaded final vendor export.
How to use this on your next episode
- Pick one real episode with a restart, a filler passage, and a pause worth keeping. Write down what you intend to remove before editing.
- After export, listen across each join without watching the transcript. Check that the neighboring answers still make sense.
- Search the exported captions for guest names, places, episode numbers, prices, and measurements. Count repairs as part of the job.
- If you compare tools, use the same source and check the final files from each. Track your own edit and repair time.
We cannot draw an overall winner or a general speed claim from this test. The Descript run included browser-agent waits and interruptions, so its elapsed wall time is not comparable editing time. Neither test establishes whether a paid plan is worth buying for your show.
Hear the restart: Watch the before-and-after clip (51-second MP4). It pairs excerpts from the synthetic source and the Descript Free audio export. It is an illustration of this one edit, not a recording of the editor interface or a listening-quality verdict.
Try the same edit
The free exercise has the unedited source audio and the three-cut brief. Use the scorecard to record your exported-audio checks, caption repairs, and time.