Audio Restoration
Audio restoration that rescues the take
Lose the hiss, echo, hum, clicks and clipping, or keep only the voice and drop everything behind it. Choose the repairs a recording needs; the clipping, click and hum repairs look for their problem first and leave the rest of the file untouched.
Uses one minute of your allowance per minute of audio: a four-minute track costs four minutes.
Hear it
Hear it on your own track, free
Free to hear, level-matched against your original. Up to 5 minutes; uploaded, processed, deleted within 24 hours. Keeping the file is a one-time purchase, with no account needed.
Drop the damaged recording, or choose it. It is uploaded to our servers, processed, and deleted within 24 hours. Up to 5 minutes.
How it works
What it measures, what it changes
And what stays yours to decide.
- 01
You choose the repairs
Steady noise, echo, hum, clicks and clipping are different problems with different fixes, so each one is its own switch, and a voice buried in noise or music has one of its own. Run only what the recording needs: the clipping, click and hum repairs search for their problem first, and a repair that finds nothing leaves the file as it was.
- 02
Each repair has its own method
Hiss and other steady noise are reduced from an estimate made on the recording itself, so there is no noise sample to select. Hum is followed as it drifts and subtracted harmonic by harmonic. Clipped peaks and clicks are redrawn from the sound on either side. Echo is reduced by predicting the room's tail from the sound before it. Clipping is repaired first, while the clipped samples still sit flat at the limit, which is how they are found.
- 03
You set how far it goes
One strength setting decides how hard the noise, click and echo repairs work, and it starts in the middle. The usual way restoration fails is not a little noise left in. It is a take scrubbed so hard the performance goes with it. Each repair is its own switch, so you can hear one on its own before running them together.
The specifics
What you would ask before trusting it with your audio
The formats, standards and numbers behind every result, each one measured.
- Hiss and steady noise
- The denoiser estimates the noise from the recording itself: where a frequency band holds a steady level, it treats that level as noise and reduces it, and where music or speech is playing, it holds back. It is made for noise that stays the same, such as tape and preamp hiss, fans and air conditioning. It never turns a band down by more than 26 dB, so pauses keep a natural floor instead of dropping to dead silence.
- A voice in noise or music
- Traffic, other voices, music, wind: noise that changes needs more than a denoiser, so Keep only the voice uses a separation model to take away everything that is not the voice. On real-world noise it left about a tenth as much noise and distortion on the voice as before (10 dB). Background to keep returns a little of the room, so the voice sounds natural.
- Clipping
- Clipped peaks are found as runs of samples stuck at the limit, then rebuilt from the sound around them, using the fact that the true peak went past the limit. The flat top is replaced by the shape the peak should have had, not smoothed over.
- Clicks and pops
- Clicks and pops up to 5 ms long are found by how sharply the waveform jumps compared with the sound just around it, so a click in a quiet passage is caught as surely as one in a loud passage. Each is redrawn from the samples on either side, and the music around it is left exactly as it was.
- Hum and buzz
- The hum repair finds the mains frequency, 50 or 60 Hz, and its harmonics up to 4 kHz, follows it as it drifts, and subtracts only the harmonics that stay steady through the recording. A bass note near 50 Hz comes and goes, so it is left alone, and even a buzz with a faint fundamental is found by its harmonics. If there is no hum, nothing changes.
- Echo and room reverb
- Echo is reduced by predicting the room's tail from the sound just before it and subtracting what that prediction explains. The first 70 ms or so of reflections are kept on purpose: they carry the body of a voice or an instrument, so the result sounds closer, never thinner.
- Files and level
- WAV, AIFF, FLAC, ALAC, MP3, AAC, Ogg and Opus, up to an hour per file. The repaired file comes back as WAV at the source sample rate. No repair adds gain: the only level change is one turn-down of the whole file, made only if a repair pushed a peak over full scale.
Questions
What people actually ask
- How do I remove background noise from a voice recording?
- If the noise changes, like traffic, people talking, music or wind, choose Keep only the voice: a separation model takes away everything that is not the voice. If it is a steady hiss or fan, Reduce noise is enough. Keep only the voice is not for songs, because it takes the music away too.
- Can you fix a clipped recording?
- Yes. A clipped peak is rebuilt from the sound around it, using the fact that the real peak went past the limit, and the rebuilt peak replaces the flat top.
- Will removing noise make my vocal sound underwater?
- That watery, chirping sound comes from turning parts of the sound down too far and too unevenly. The denoiser never turns a band down by more than 26 dB, and it smooths its changes across neighboring moments and frequencies, which are the two usual causes. If you still hear it, lower the strength.
- Should I restore before or after mixing?
- Before, on the separate tracks, where you can. Once a noisy track is mixed in, its noise sits under every other instrument, and taking it out of the mix takes some of them with it.
- How do I get rid of room echo?
- Choose Reduce echo. It lowers the room's tail and keeps the first reflections, so the voice sounds closer and keeps its body. For a voice in a busy room, Keep only the voice takes the background away too.
Try it
Put your own track through it
The free plan runs the whole engine on your whole recording and plays you the result, matched for loudness against the original. No card. You pay when you want the full-quality file, one track at a time, or with a plan if you make a lot of them.
Who it’s for