This is a copy. Use Snipmill at snipmill.com.

Skip to content

An AI takes the noise out from behind a voice, then the speech is made clearer and the loudness even, on Snipmill's server.

Settings

Settings

NoiseBehind a voice; not for music
VoiceCuts rumble, lifts speech
LoudnessWhole file at −16 LUFS
Runs on Snipmill's server: only the sound is sent, and deleted after. How we keep it private

Good to know

The AI noise remover, and where it runs

Noise is taken out by an AI model on Snipmill's server. It keeps the voice and takes out what isn't speech. There is no simpler fallback: filters that only turn down steady noise leave most of it, so when our server can't be reached, the tool doesn't run.

Where it runs
What to expect: By an AI model on Snipmill's server, on a computer, phone or tablet
What is sent
What to expect: The sound alone, 48 kHz mono AAC at 64 kbps, when you press Start; a video's picture isn't sent
Other voices or music behind the speaker
What to expect: Turned down, but not always removed: play the result
Sound it keeps
What to expect: 48 kHz, in mono; the cleaned sound comes back as Opus at 64 kbps

What noise removal does to your recording

The noise is taken out on Snipmill's server; everything else runs in your browser. In order:

Noise
What happens: With Noise on, the sound goes to Snipmill's server, where the AI model takes out what isn't the voice; the cleaned sound comes back and the server deletes what was sent
Low rumble
What happens: Everything under 80 Hz is cut (with Voice on Clearer)
Voice
What happens: With Clearer: gentle compression, so loud and soft words come closer, and +3 dB around 3 kHz
Loudness
What happens: With Even: −16 LUFS for the whole file, as one steady change worked out from a measuring pass, with peaks under −1.5 dBTP
Length, mono or stereo
What happens: The length is kept; with Noise on the result is mono
Format
What happens: Your file's own: MP3 stays MP3, M4A and AAC come out as M4A, WAV (16-bit), FLAC and OGG stay; Opus comes out as OGG (Vorbis); anything else as MP3. MP3 and M4A at 192 kbps
Video
What happens: The picture is copied as it is when it is H.264 or HEVC, else encoded again as H.264; the sound goes back in as AAC at 192 kbps, in an MP4

Which setting to pick

Three switches, all on at the start:

  • Noise, Remove or Keep: Remove runs the AI noise remover on the whole file. Keep it for a recording without noise, where only the voice or the level needs work.
  • Voice, Clearer or As is: Clearer cuts rumble under 80 Hz, brings loud and soft words closer and adds +3 dB around 3 kHz.
  • Loudness, Even or As is: Even sets the whole file to −16 LUFS, the level Apple asks podcasts to keep (it writes it as −16 LKFS, the same scale). Turn it off to keep a recording's own level.

What people clean

  • A 3-minute voice memo recorded next to an air conditioner, as M4A: the hum is taken out from behind the voice, and it comes back as an M4A of about 4.2 MB.
  • A 20-minute phone video of an interview with traffic outside, cleaned on a laptop in Chrome: the traffic is taken out, the H.264 picture is copied as it is, and the voice comes out at −16 LUFS.
  • A vlog recorded in a busy market: the crowd behind the voice is turned down, but other voices are speech too, so some can stay; play the result before you post it.

What goes to our server

Only the sound of each file goes to Snipmill's server, when you press the button that starts the job; the line under it says so. The server deletes it once it has answered and keeps no copy. With Noise on Keep, the sound isn't sent. The built-in copy of FFmpeg that shapes the voice and the loudness, and makes the file you save, runs in your browser; a video's picture isn't sent.