NAUPLY

Repair AI tutorial: fix damaged audio, module by module

4 min read · Updated

Repair AI fixes damaged audio: clipped peaks, clicks and crackle, mains hum, hiss and background noise, room reverb. Put it as an insert on the track to repair, before the rest of the processing. The numbers match the markers on the screenshot.

Step by step

NAUPLY Repair AI tutorial: fix damaged audio, module by module interface, with numbered markers for each step

  1. Open the TEMPS RÉEL tab

    Insert Repair AI on the damaged track (voice, podcast, field recording, vinyl or tape rip, instrument) and open its window on the TEMPS RÉEL (realtime) tab.

  2. Start from a preset

    Pick a preset close to the problem: Voice & Podcast > Podcast Room Noise, Music Restoration > Vinyl Rip, Location & Film > Field Recording Hum 50 Hz…

  3. Switch on only the modules you need

    Or switch ON only what you need, in rack order: 1 De-clip, 2 De-click, 3 De-hum, 4 Denoise, 5 De-reverb.

  4. Learn the noise (Denoise)

    During a passage of noise only, with no voice or music, switch LEARN on and then off: PROFILE goes to Learned and the noise print is saved with the project. Keep REDUCTION moderate for a natural sound.

  5. Listen to what is removed

    A module's LISTEN plays only what it takes out. If you hear the voice or the music in it, turn that module down.

  6. Compare and set the output

    Compare with BYPASS: latency is kept, so there is no offset in A/B. Then set OUTPUT.

De-clip: rebuild clipped peaks

When a recording clipped, the tops of the waveform are flattened. De-clip rebuilds them. In AUTO it finds the clipping level by itself (flat tops); switched off, it uses LEVEL, useful for soft clipping or a file that was already re-encoded. AMOUNT sets how much is rebuilt.

Rebuilt peaks go above the old ceiling: HEADROOM lowers the output so they do not clip again.

De-click: clicks, crackle, mouth noises

De-click removes clicks and crackle (vinyl, digital dropouts, mouth noises) without touching drum hits. Higher SENSITIVITY catches weaker clicks, but set too high it also touches transients. MAX LENGTH sets the longest a click can be: anything longer is left alone, which protects note attacks.

De-hum: mains hum

MAINS on Auto detects 50 or 60 Hz; you can also force 50 Hz or 60 Hz. For a buzz at another frequency, play a passage where only the buzz is heard and press LEARN: two seconds are analyzed and MAINS switches to Learned.

HARMONICS sets how many harmonics are removed (50, 100, 150 Hz…): the more, the more high buzz disappears too. WIDTH sets the width of each notch: narrow leaves voice and music untouched; wide follows unstable hum better.

Denoise: hiss and background noise

REDUCTION sets the maximum noise reduction: what is left keeps a natural floor. On PROFILE Auto, the plugin follows the noise by itself in quiet passages; on Learned, it uses the print learned with LEARN, more precise when the noise is steady.

Higher SENSITIVITY treats more material as noise (more effective, but it may damage the signal); higher SMOOTHING gives a calmer residue, with fewer metallic "little notes", at the cost of some precision on transients.

De-reverb: shorten the room tail

REDUCTION sets the maximum reduction of the late reverb. TAIL matches the length of the room's tail: set it until the tail disappears without pumping. In AUTO, the plugin estimates it on phrase endings, as long as there are pauses.

Spectral repair: erase a cough, a ring, a squeak

The RÉPARATION SPECTRALE (spectral repair) tab works on a clip: drop a file on the spectrogram (or OPEN FILE), or start playback in your DAW, choose a length from 5 to 60 seconds and click CAPTURE. The realtime chain is not affected.

Pick a tool (RECTANGLE for a time and frequency area, LASSO for a free shape, TIME for all frequencies, HARMONIC for a note and its multiples) and draw around the noise. PLAY SELECTION checks the selection, then REMOVE rebuilds it from its surroundings. ATTENUATE lowers it by a number of dB, REPLACE swaps in the passage just before or after, DE-TONE removes a steady whistle.

PLAY REMOVED plays what was taken out, UNDO / REDO keep the history. Drag the DRAG TO DAW handle onto an audio track, or use EXPORT WAV (24-bit, same sample rate and length).

For spoken word

For a podcast or a voice-over, Clarea is built for the voice: neural denoising, breaths, plosives and a leveler. And when noise only bothers you between phrases, a noise gate like Gatea is often enough.

Frequently asked questions

In what order should Repair AI's modules be used?
In rack order: De-clip, De-click, De-hum, Denoise, then De-reverb. Each repair prepares the next: denoising a clipped or click-ridden signal gives worse results.
Can Repair AI be used live?
No: its fixed latency (about 100 ms at 48 kHz) is compensated by the DAW on playback, but it makes it unsuitable for live monitoring.
Is my audio sent anywhere?
No. All processing happens on your computer.

NAUPLY news

Plugin releases, updates and offers. A few emails a year, one-click unsubscribe.

More guides