We tested auto-editor's silence cutting on three dialogue clips
auto-editor removes the quiet parts of a video by loudness. We ran it on three 60-second clips from the open film Tears of Steel, at its default setting and a stricter one, and measured what it cut.
Tested by us on Nov 5, 2026, on the clips and machine listed below. Every number on this page was measured in that run.
The result
On film dialogue with a music score underneath, auto-editor's default setting barely cut anything: 1.33 seconds out of 180 seconds across the three clips, and nothing at all on two of them. The music keeps the sound level above its silence line, so there is little it sees as quiet.
The stricter setting (audio threshold 8%) cut more, from 0.71 to 5.42 seconds per clip, but some of what it removed was speech: up to 1.71 seconds of the time the official subtitles show someone talking, on the busiest clip. It also left very short pieces, some under half a second long, which play as jumps.
So for a talking head with a clean, quiet room, auto-editor is worth trying. For anything with a music bed or room noise, check every cut by eye, or remove the music first.
What failed or surprised us
- Default setting: on two of the three clips it cut nothing at all, because the score never drops below its threshold.
- Stricter setting: it cut into speech, up to 1.71 seconds of subtitle-marked dialogue on one 60-second clip.
- Stricter setting: it left kept pieces as short as 0.46 seconds mid-clip (0.42 seconds at the very end of tos_b) and removed gaps as short as 0.04 seconds, which show as visible jumps.
- On its first run auto-editor downloaded its own program file from its official GitHub release; plan for that on a machine without internet access.
What we tested and how
- Tool: auto-editor version 29.3.1, Unlicense.
- Settings compared: Default setting; Stricter: --edit audio:threshold=8%.
- Input clips: Tears of Steel, (CC) Blender Foundation, mango.blender.org, CC-BY 3.0. 60 s excerpts re-encoded to H.264 720p.
tos_a: 60 s starting at 20 s of the film, SHA-256945262dcbe106f1e...tos_b: 60 s starting at 180 s of the film, SHA-2564097773b615cddfa...tos_c: 60 s starting at 360 s of the film, SHA-2563d3a8cd015310349...
- Machine: AMD Ryzen 7 7735U (8 cores, 16 threads), integrated Radeon 680M, not used (CPU-only runs), 31.2 GB RAM, Windows 11 10.0.26300. The laptop was shared with other work, so timings are what a busy machine gives, not a clean benchmark.
- Runs: one cold run, then three timed runs per clip and setting. Run on 2026-11-05.
auto-editor finds quiet parts by loudness and cuts them out. We ran its default command, then the same command with a stricter audio threshold, on the same three clips. To judge whether speech was cut, we used the timings of the film's official English subtitles as an approximate map of when someone is talking.
Measurements by clip
Default setting
| Clip | Output length (s) | Cut (s) | Cut (%) | Speech in subtitles (s) | Of that, cut (s) | Median run (s) | Peak memory (MB) |
|---|---|---|---|---|---|---|---|
| tos_a | 58.67 | 1.33 | 2.2 | 25 | 0.46 | 26.4 | 490 |
| tos_b | 60 | 0 | 0 | 27.2 | 0 | 38.2 | 493 |
| tos_c | 60 | 0 | 0 | 24.5 | 0 | 26.1 | 482 |
Stricter: --edit audio:threshold=8%
| Clip | Output length (s) | Cut (s) | Cut (%) | Speech in subtitles (s) | Of that, cut (s) | Median run (s) | Peak memory (MB) |
|---|---|---|---|---|---|---|---|
| tos_a | 57.92 | 2.08 | 3.5 | 25 | 0.58 | 23.6 | 487 |
| tos_b | 54.58 | 5.42 | 9 | 27.2 | 1.71 | 33.7 | 487 |
| tos_c | 59.29 | 0.71 | 1.2 | 24.5 | 0 | 34.8 | 482 |
Reproduce it
auto-editor clip.mp4 -o out.mp4auto-editor clip.mp4 --edit audio:threshold=8% -o out.mp4Install auto-editor with pip, then run one of these on your clip. We exported the cut list with --export v3 to measure where it cut.
Method and limits
- Three 60-second clips from one film are a narrow test. A clean talking-head recording would behave differently, and that is our next silence test.
- Subtitle timings are not exact speech boundaries; a subtitle line can start before or end after the words. Treat the speech-cut figure as approximate.
- Timings include writing the output video. The laptop was busy with other work, so the times show what a loaded machine gives, not the tool's best speed.
- We did not measure time saved: that would need the same clips cut by hand as a baseline.
- Footage: Tears of Steel, (CC) Blender Foundation, mango.blender.org, used under CC-BY 3.0. We do not publish the cut videos.
Page checked on Nov 4, 2026.