AI Music Cleaner for Suno and Udio Songs

Check the sliders under the player. If the top of your file is intact (MP3 from 128 kbps up, WAV and FLAC — the usual export of AI generators), there is nothing to restore, and only Brightness and Reduce harshness change the sound; if both are at zero, the tool says so and does not count the run. Brightness is easiest to hear on cymbals, consonants and the "air" of a track.

Move Brightness under the player to the left, to −4…−6 dB. The network generates the "sand" together with the music, in the same frequencies as cymbals, consonants and "air", so no method removes it separately from them: we tested neural restoration, splitting into stems and denoising the vocal alone — the sand only went away together with the top end. At the same time, the 8–16 kHz band of AI tracks is on average 5–6 dB louder than in mixed recordings of real musicians (measured on 42 and 48 tracks). Brightness at −6 dB takes about 5.5 dB off the 8–12 kHz band without touching the mids and brings the top back to the level of real recordings — the sand gets quieter with it. If particular spots sound harsh, add Reduce harshness.

Apollo is a neural network published at ICASSP 2025 (Li et al., "Apollo: Band-sequence Modeling for High-Quality Audio Restoration"). It was trained to restore MP3-compressed music — specifically MP3s in the 24-128 kbps range — toward lossless quality. On a file whose top is not cut off it returns the input almost unchanged, so here it runs only where there is something to restore.

That lift of the upper mids came not from the neural network but from the filter joining its output to the original: on a file with an intact top, the model returns its input almost unchanged. Now the model runs only where the top is cut off, and the old sound became the Brightness slider: by default it sits at the same setting (+7.5 dB around 10 kHz), takes seconds instead of minutes, and does not overload the peaks.

Apollo is trained on music. For speech cleanup (interviews, podcasts), use our Audio Enhancer or Voice Enhancer tools — they target speech-specific noise patterns.

The engine and the track's length. Quick mode takes seconds on any device. Deep mode works through the song from start to finish and does the same amount of work per second of audio: minutes on a computer with a graphics card, several times longer without one — the estimate for your device is on the button before you start. A file whose top is intact is checked in seconds: the model does not run for it.

Music whose top compression cut off: MP3 up to 96 kbps, recordings with a low sample rate, old files from the web. The tool measures the cut itself: below roughly 16 kHz there is something to restore, and the lower the cut, the more noticeable the result. Files without a cut — MP3 from 128 kbps up, WAV and FLAC — have nothing to restore: their sound is shaped by the Brightness and Reduce harshness sliders. Heavily damaged material (clipping, dropouts) is a job for the Audio Restoration tool.

No. The restoration runs in your browser — your audio file is not sent to us. We log only anonymous performance events (model load time, inference duration) to spot bugs and improve speed; see the Privacy Policy.

With any file whose top compression cut off, whichever generator made it. The generator itself does not matter: the tool measures the cut in the file. If the top is intact, there is nothing to restore, and the sound of such a song is shaped by the Brightness and Reduce harshness sliders.

Not MP3 again, if you can avoid it. The whole point of the run was to undo what MP3 compression did, and saving the result back to a lossy format immediately spends part of what you recovered. WAV or FLAC keeps it whole for further work or for archiving; a lossy export makes sense only at the very end of the chain, when the file is going straight to a listener and nothing else will touch it.