Auto cut audio online

It measures loudness, not words. FFmpeg's silencedetect filter scans the waveform and flags every stretch that stays below a dB threshold for at least 0.8 seconds. Those stretches get removed, minus 150 ms of breathing room kept at each end. No transcription, no AI model, no upload of your text anywhere.

Not at the default settings, and the reason is arithmetic. A gap has to stay quiet for 0.8 s to register at all — no syllable-length pause qualifies — and we then hand back 150 ms at each end, so the cut starts 150 ms after you stopped talking and ends 150 ms before you start again. Push the threshold above roughly −25 dB and that protection stops mattering: at that point your quieter words are themselves below the threshold. The slider above shows exactly where each file crosses that line.

Because your file probably did not have much dead air in it. Time removed is not a property of the tool, it is a property of your recording. Every one of the five files we benchmarked was an already-published, already-edited recording, and the tool removed between 0% and 3.3% from them at default settings — a human editor had taken the pauses out already. A raw one-take with real thinking time in it has far more to give back. If you think your file has more, raise the threshold a few dB.

Then your noise floor is too close to your voice. Our auto-calibration sets the threshold 5 dB below the file's mean volume, and if constant hiss, air conditioning or room tone sits above that line, the detector never sees a quiet stretch. We hit this in our own bench: on an Apollo 11 comms tape the tool removed 0.01 s out of 5 minutes, because tape hiss fills every pause. Fix: drag the threshold up a few dB — on our long-talk file, 5 dB took it from 4.2% to 15.4%. If no setting gives a sensible result, the problem is noise, not pauses. That tape went from 0% at −28 dB to 90% at −27 dB with nothing usable in between. Denoise first.

No. An “um” is sound, not silence, so a loudness-based detector keeps it. Removing fillers requires transcription — Descript and Adobe Podcast do that, we don't. If your problem is verbal tics rather than dead air, we are the wrong tool and one of those is the right one.

Yes, it is re-encoded, and we would rather say so than claim otherwise. Every audio export is rebuilt with LAME at quality 2 (VBR, roughly 190 kbps) and hands you an MP3 regardless of what you uploaded. For speech that is transparent in practice, but it is a generation of lossy encoding, and if your source was a 64 kbps file the output can end up larger than the input — in our bench, a 2,344 KB clip came back as 3,155 KB. If you need a bit-exact WAV or FLAC master, do the cut in Audacity or Reaper instead.

No — not on the free tier, not anywhere. There is no audio stamp, no appended tag, no trailing jingle. The output is your file minus the gaps.

5 minutes per file, 200 MB, one file per day, no account needed. Pro raises that to 60 minutes and 2 GB with no daily cap. The 5-minute ceiling is genuinely tight — shorter than several manual trimmers allow — so if you are here to tighten a 40-minute episode, know that up front rather than after the upload bar finishes.

Sort of, and not the way you probably mean. Drop an MP4, MOV, MKV, WEBM or AVI and we cut video and audio together, in sync, and give you back an MP4 — not a standalone MP3. There is no video-to-audio extraction path today. mp3cut and Clideo both do that conversion, and for that job they are the better tool.

All of them, because it never listens for words. Silence detection is a loudness measurement, so Japanese, Arabic, Wolof and English all behave identically. Only your microphone and your room change the result.

It goes to our storage, gets processed, and is deleted an hour later by a cleanup job that sweeps every 10 minutes. No training on your audio, no human listening to it, no account required for your first file. Being straight about it: this is weaker than a tool that never uploads at all. Browser-only cutters exist and if the recording is sensitive, that is a real argument for using one — they cannot process a 200 MB file the way we can, but they never receive it either.

Whenever the edit is one cut you can see. Topping and tailing a track, making a 30-second ringtone, grabbing one quote out of an interview — mp3cut, AudioTrimmer or Clideo will beat us on all of those, because dragging two handles is faster than tuning a detector. We are only worth it when the pauses are too many to click through by hand.