Audio finishing toolkit for AI-generated music.
Loudness normalization. Stem separation. Surgical region editing. Platform mastering export. Built specifically for the problems AI music generators leave behind.
Batch-normalize to Spotify, CDBaby, YouTube, Apple Music, or DistroKid specs. 16/24/32-bit output, TPDF dither on 16-bit, correct sample rate conversion. Set it once, apply to everything.
Powered by Facebook's Demucs (htdemucs model). Splits vocals from music so you can process each independently. Clean only the vocals — the music passes through untouched. Prevents the "underwater effect" you get from processing the full mix.
Load any track, mark problem sections, auto-detect artifact regions by spectral analysis. Set a different cleaning preset per region. Preview each fix before committing. Build, export, done.
DeepFilterNet neural denoiser and WPE dereverberation run on isolated vocal stems after separation — never on the full mix. Helps with noise floor and room echo on recorded vocals or voice-forward AI tracks.
Targeted EQ cuts for the known Suno artifact frequency zones: 320–400 Hz mud, 3.5–5 kHz metallic ring, 5–7 kHz AI shimmer, 8–12 kHz v5.5 hiss. Tunable Q and gain per notch. Spectrogram view helps you find your own.
Drop a folder of tracks, pick a preset, walk away. Step Mode lets you compare Gentle → Balanced → Strong on your originals before committing. Sample clip generator cuts preview MP3s from every queued file automatically.
Most "AI artifact remover" tools oversell. We'd rather be straight with you.
The metallic harmonic artifact that defines Suno v5.5 is baked into the synthesis model's transposed convolution layers. The artifact frequency and the real musical note underneath it are literally the same Hz. A notch filter that kills the ring also softens the note. CleanWave's notch system reduces it — narrow, shallow cuts that minimize collateral damage — but it cannot remove it cleanly without affecting the music. No tool can. This is a problem for Suno to fix at the model level.
It was trained on human speech recordings. Running it on a full music mix will destroy the music. CleanWave only applies it to isolated vocal stems after Demucs separation. On voice-forward AI tracks it helps; on heavily produced instrumental content it adds little.
Demucs is the best freely available separator, but bleed exists. Heavily reverbed vocals, complex polyphonic passages, and certain instruments can end up in the wrong stem. The stems are good enough for cleanup purposes; they're not studio-quality isolation.
CleanWave is genuinely useful for loudness normalization, mastering export, and reducing some artifacts on some tracks. If a track sounds good already, it makes it platform-ready. If a track has heavy AI artifacts, it reduces them without making things worse — which is more than most paid tools do.
Free. No account. No telemetry. Everything stays on your machine.
Or grab the portable zip and run install.bat yourself:
CleanWave_Setup_1.0.exe (or extract the zip)run.bat