AUD · Audio tools

Spectral Repair

SpectrogramHz / time
Load a file to see its spectrogram.

Shareable link

Settings are written to the URL as you change them. Nothing differs from the defaults yet.

Oxyd To put the result on a track with other audio, Oxyd is a multitrack editor with the same effects. Open Oxyd

Some noises are not noise. A cough in a lecture, a chair in a string quartet, a phone in an interview: each sits in one place on the spectrogram, and a denoiser, which subtracts what is steady, leaves them alone. This takes them out one at a time.

Draw the box

Load the file and the spectrogram appears under the waveform, time left to right and frequency bottom to top on a log scale. Drag over the mark to draw a box. Click a box to select it, press Delete or the button to remove it, and click anywhere else to move the playhead. Up to 64 boxes are kept in the link, so a repair survives a reload and can be handed to the workflow.

Fill or attenuate

Fill replaces every bin inside the box with the average that bin carried over the Context window before the box: 800 ms by default, longer for a steadier fill, shorter to follow a texture that moves. The bin keeps its own phase, so the frames still overlap coherently and the edges of the box are invisible on room tone.

Attenuate turns the box down by the Attenuation amount instead. It keeps whatever shape was inside, which suits a sound to keep that overlaps the one to remove.

Both modes feather the box over 20 ms in time and two bins in frequency, so an edge is a fade rather than a click.

Where it stops

A cough over speech cannot be replaced with the speech that was under it; the fill restores the room and the words stay missing. A box drawn over a whole band for a whole phrase is an equaliser with extra steps, and the denoiser or the hum remover is the better tool for anything steady. For a click shorter than a few milliseconds, the declicker finds and interpolates it without a box.

Frequently Asked Questions

A rectangle on the spectrogram: a span of time and a span of frequency. A cough is half a second wide and reaches from 200 Hz to 3 kHz; a phone ring is a thin band at one pitch; a creak is a short broadband streak. Draw the box over the mark and nothing outside it is touched.

It replaces each bin inside the box with the average that bin carried just before the box opened, for as long as the Context control says. On room tone, a pad or a held note the fill is indistinguishable from the surroundings. On speech it fills the gap with the room, not with the words that were under the cough.

When the sound to keep is louder than the sound to remove and shares its band, such as a cymbal under a chair creak. Fill would replace the cymbal with the average cymbal; attenuation turns the whole box down and keeps its shape. Try 12 to 20 dB first.

A fill from both sides would need to hear the end of the box before writing its start, and a box can be seconds long. The preview would start that late. Filling from the past keeps one window of delay, the same as every other spectral step, and on steady material the two are the same fill.

Yes. Each channel keeps its own running texture, so a stereo room fills with a stereo room. It reads WAV, MP3, M4A, AAC, Ogg, Opus, FLAC, AIFF and WebM audio, and exports in the format it was given.

Explore Our Tools

Browse all tools