The take happened once. The fridge, the fan and the traffic happened too. PaperEdits names the exact noise in your file, picks the one fix that works on it, and shows you the silence.
Made for: YouTube voiceovers, interviews, phone footage, screen recordings
“The chain is squeaking through the whole video. I didn’t notice until I got home.”porchlight_vids
“I ran noise reduction and got a little improvement. I want much more.”garage_don
“Whenever I use a noise removal tool, it makes my voice sound weird.”nadia.vlogs
The problem with every noise slider: it’s one tool for a dozen different noises. A hum, a keyboard and a crowd need completely different fixes, and the slider doesn’t know which one you have.
Drop your video in. It starts diagnosing the moment the audio loads, no waiting for a transcript.
Read the plain answer: “Found 50 Hz electrical hum”, or clicks, hiss, chatter. Multiple at once is normal.
One click fixes it. Hold A/B to hear the original any time, and undo if you change your mind.
The right tool per noiseHum gets filtered out instantly. Keyboards, hiss and chatter go to a smart model that sliders can’t match.
Your voice stays yoursWe transcribe before and after in testing: the words come back identical. No robot voice.
It warns before it hurtsIf the background is loud enough that cleanup could thin your voice, it tells you before you export.
Free to startA free account, no upload, no watermark. It all runs in your browser.
No. Unlike tools that rebuild your speech, this removes noise around your voice. We test by transcribing before and after: the words come back identical.
Because a generic slider is usually the wrong tool for your specific noise. A keyboard or a crowd can’t be filtered out by anything fixed. PaperEdits figures out what your noise is first, then applies the fix that actually works on it.
That’s a different problem with its own tool: separate overlapping voices.
Diagnosis reads the quietest 50 ms of every second: level plus variability. Hum is notched at the fundamental and three harmonics, 13 dB down in one click. Everything else goes to GTCRN, a 192 KB model in a WebAssembly worklet. On a real air conditioner mix the floor drops from −27 to −72 dB with the voice held within 0.4 dB. The test files are public.