Remove silence in Audacity with Truncate Silence
Audacity’s Truncate Silence effect shortens pauses for free — if you set it correctly. Get the threshold, duration and “truncate to” wrong and the result sounds robotic and gasping. Below: the exact settings for voice, the mistakes that ruin it, and a no-install browser alternative for when you just want the result.
The three settings that decide everything
Truncate Silence lives under Effect > Special > Truncate Silence. It has three controls that matter, and each one has a failure mode. Here’s what each does and where to put it for spoken voice.
| Setting | What it does | Recommended for voice |
|---|---|---|
| Threshold | The level below which audio counts as silence. Too high and it eats into quiet speech; too low and it ignores real pauses. | Set it just above your room's noise floor. On a treated recording that's often around −45 to −40 dB; on a noisy laptop mic you may need −35 dB or the effect finds nothing. |
| Duration | The minimum length a quiet stretch must last to be treated as a silence to shorten. This is the guard against chopping natural micro-pauses. | 0.4–0.6 s keeps the rhythm of speech. Drop below ~0.3 s and it starts removing the breaths and beats that make talk sound human. |
| Action → Truncate to | How much silence to leave behind. This is the setting people get wrong most often — set it to 0 and every join sounds hacked. | Leave 0.2–0.3 s of air. A 0.5 s target is safe; 0.1 s or less is where the robotic, gasping sound comes from. |
These are starting points for a typical spoken-voice recording, not universal constants — your room and mic set the right threshold. The official effect reference is on the Audacity manual.
Step by step, without the robotic sound
1. Open Truncate Silence
Select the whole track (Ctrl+A), then go to Effect > Special > Truncate Silence. Working on the whole track keeps the effect consistent across the recording.
2. Set the Threshold to your noise floor
Start around −40 dB and adjust: if it removes nothing, raise it toward −35 dB; if it eats quiet words, lower it toward −45 dB. There is no auto-detect — you're reading your own file.
3. Set the minimum Duration
Use 0.4–0.6 s so only real pauses are shortened and natural micro-pauses survive. This is what stops the result sounding clipped.
4. Choose Truncate to, and leave air
Pick “Truncate Detected Silence” and set Truncate to 0.2–0.3 s. Do not set it to 0 — that's what creates the robotic, breath-less sound.
5. Preview, then apply and export
Press Preview and listen to two or three joins before committing. If it sounds hacked, raise Truncate to. Then apply and File > Export your audio.
The four mistakes that wreck the result
When Truncate Silence produces choppy, unnatural audio it’s nearly always one of these — and each has a specific fix.
Truncate to 0 → robotic, gasping speech
The single most common mistake. Removing every millisecond of pause welds words together and clips breaths mid-inhale. Speech has rhythm; strip all of it and the ear immediately hears “machine”. Leave 0.2–0.3 s and it sounds edited, not processed.
Threshold too high → it cuts into your voice
Set the threshold up near the level of quiet speech and Truncate Silence treats the soft ends of words as silence and shortens them. You get clipped word tails and a lisping, choppy quality. Lower the threshold until only the true gaps are caught.
Duration too short → natural pauses vanish
Below ~0.3 s the effect attacks the small beats between phrases, the ones that carry meaning and let a listener breathe. The recording ends up unnaturally fast and hard to follow. Keep the minimum duration at 0.4 s or more for voice.
One global setting for a whole varied recording
Truncate Silence applies one threshold across the entire file. If your levels drift — you leaned in, the mic gain changed, a guest is quieter — a setting that's right for one section is wrong for another, and you get inconsistent cuts. Per-section fixes mean re-running it on selections, by hand.
Audacity Truncate Silence vs Make Autocut
Audacity is a full, free audio editor and this is a single-purpose silence cutter — so this isn’t “better”, it’s a different trade. Audacity wins on scope and being free with no limits; the web tool wins on not having to find the right threshold yourself.
| Audacity (Truncate Silence) | Make Autocut | |
|---|---|---|
| Silence detection | You set the threshold and guess-and-check | Auto-calibrated per file to your noise floor (mean − 5 dB) |
| Install | Desktop app, download and install | Runs in the browser, nothing to install |
| Padding / keep-air control | “Truncate to” — one global value | Padding per join (default 150 ms), plus a minimum-silence length |
| Consistency across a drifting recording | One global threshold; re-run per selection by hand | Reads the whole file's floor, one pass |
| Works on video | Audio only — you'd strip the audio first | Yes — MP4/MOV in, cut video or a timeline out |
| Noise reduction, EQ, normalize, mixing | Yes — a full audio editor | No — silence only, by design |
| Cost | Free and open source | Free tier (no account), then Pro |
Where Make Autocut auto-calibrates: it reads your file’s mean volume and sets the cut threshold to mean − 5 dB, keeps a 0.8s minimum silence and 150ms of padding at each join — the equivalents of Audacity’s Threshold, Duration and Truncate-to, chosen from your file instead of by you.
Best of both: clean in one, cut in the other
These tools don’t compete — they sit at different points in post. Audacity does the things a silence cutter never will: noise reduction, EQ, de-essing, normalization, mixing multiple tracks. A silence cutter does the one thing that’s tedious to do by hand across a long recording.
Master first, then cut
Run noise reduction and EQ in Audacity, export a WAV, then send it through Make Autocut for a fast, calibrated silence pass. Cleaner input means the detector’s job is easier, because the noise floor is lower and better defined.
Cut first, then finish
Or cut the silence here, download the file, and open it in Audacity to mix in music, level, and master. The silence pass is done and consistent, so you’re finishing a recording that’s already the right length.
For podcasters: cut the dead air, keep the rhythm
A one-take conversation is where silence removal earns its keep — hundreds of small gaps while people think, breathe, or find the next point. But a podcast lives or dies on rhythm, so the goal isn’t to remove all the pauses; it’s to remove the dead ones and keep the ones doing work.
On our benchmark (2026-07-16), unscripted spoken content gives back real time: a long-form talk lost 2.9% of its length across 10 cuts, a recorded lecture 1.4% — while a tightly prepared voice-over only gave back 3.3%, because there was barely any dead air to begin with. The rawer the recording, the more there is to cut. That’s exactly why keeping a 0.8s minimum silence and 150ms of padding matters: it takes out the long dead gaps and leaves the natural half-second beats that keep speech human.
The full podcast workflow — from raw recording to a tightened episode — is on the podcast editing page, and the auto cut audio benchmark has all 155 measurements on downloadable files.
Want the result without the settings?
First file is free, no account, no install. It reads your file’s noise floor and shows the cut on the waveform before you commit — no threshold to guess, no robotic joins.
Audacity silence removal: common questions
Is Make Autocut free, like Audacity?
▾
The free tier is real — one file a day, up to 5 minutes and 200 MB, no account, no watermark. Audacity is free and open source with no limits at all, which is a genuine advantage if you want everything on your own machine. The difference isn't really price; it's that Audacity makes you set the threshold and check it yourself, and this calibrates to your file automatically. For longer recordings past 5 minutes you'd move to Pro (60 min).
Does Make Autocut need an install like Audacity?
▾
No — that's the main practical difference. Audacity is a desktop app you download, install and update; Make Autocut runs in a browser tab, so it works on a locked-down work laptop or a Chromebook where you can't install anything. The trade is that it needs a connection to upload and process, whereas Audacity works fully offline once installed.
Does either one degrade the audio quality?
▾
Truncate Silence in Audacity edits the samples in place without re-encoding, so the audio itself isn't recompressed — quality loss comes only from bad settings (clipped words, hacked breaths), not from the effect. Make Autocut's file export re-encodes (LAME −q:a 2, transparent for speech); if you want zero re-encode, take the timeline export instead and apply the cut in an editor. Both preserve quality when used right.
Does Make Autocut work on video, or only audio like Audacity?
▾
Both. Audacity is an audio editor — to use Truncate Silence on a video you'd extract the audio, process it, then re-sync it, which is fiddly. Make Autocut takes MP4 and MOV directly, cuts the video and audio together, and hands back either a cut video or an editable timeline for your NLE. If your source is a video, that saves the extract-and-resync dance entirely.
Can I use both together?
▾
That's often the best workflow. Clean the audio in Audacity first — noise reduction, EQ, normalize, the things a silence cutter doesn't do — export a WAV, then run it through Make Autocut for a fast, calibrated silence pass. Or reverse it: cut the silence here, then finish and mix in Audacity. They don't compete; one masters the sound, the other removes the dead air.
Why does my Audacity result sound choppy or robotic?
▾
Almost always the “Truncate to” value is too low — often 0. Removing every scrap of pause welds words together and clips breaths, and the ear reads that as machine-processed. Raise Truncate to 0.2–0.3 s and lift the minimum Duration to 0.4 s or more so natural micro-pauses survive. If it's still hacked, your threshold is probably too high and it's cutting into the voice itself.
Skip the threshold hunt
5 minutes, free, no watermark, no install. Auto-calibrated to your file.
Try the auto silence cut