A freeware desktop application that analyzes MP3 files and attenuates the inaudible spread-spectrum watermark that Suno embeds in its generated tracks. Built in Rust with egui/eframe, it ships as a native, self-contained binary — no runtime, no cloud processing, no telemetry.
Closed-source freeware. The signal-processing method is proprietary know-how. Publishing the implementation would let Suno (and other AI music platforms) patch the weakness it exploits. The binary is distributed for free; donations are appreciated.
Main window — load an MP3 and scan it for the embedded spread-spectrum watermark.
Detection result — high-confidence score, spectrum analyzer, and the Clean Watermark button.
Spectrogram view — the 18–24 kHz band highlight and the psychoacoustic masking model (ISO 226) used to target attenuation without audible artifacts.
A full walkthrough of the tool in action — loading a track, detecting the watermark, and cleaning it.
The video is also available in the repository:
screenshots/fullscreen.mp4.
When Suno renders a track, it imprints a low-level, broadband, pseudorandom signal onto the audio — a spread-spectrum watermark designed to survive MP3 encoding, lossy compression, EQ, and casual editing. This tool applies a series of proprietary signal-processing steps to attenuate that embedded signal and reduce its detectability.
It does not guarantee complete removal — the watermark is robust by design — but it can often fade the signal enough to lower the detection confidence score below detector thresholds.
The internal method is deliberately not documented publicly. At a high level, the tool:
- Analyzes the loaded MP3: decodes it, computes a real STFT spectrogram, extracts spectral features (centroid, rolloff, flatness, zero-crossing rate, RMS, peak), and computes psychoacoustic data (ISO 226 equal-loudness contours, simultaneous masking threshold).
- Scans for a watermark-style signature in the ~18–24 kHz band and reports a detection confidence score, combining the acoustic high-frequency signal with an AI-provenance scan of the file's metadata.
- Cleans the file by re-encoding it with a pure-Rust MP3 encoder
(
rusty_mp3), producing a brand-new*_clean.mp3with zero metadata — no ID3 tags, no encoder provenance, no "Produced with Suno" markers.
The exact combination of techniques is the proprietary know-how. If it were published, Suno's engineering team could adjust their embedding parameters to close the window — which is why the source code is not available.
The binary is fully self-contained: decode, analysis, spectrogram, and encoding all happen in-process with pure-Rust crates. There is no ffmpeg dependency, no system audio tools, no runtime Python. The only shared libraries linked are glibc/libm/libgcc, which exist on every Linux system.
- Not 100% effective. Suno's watermark is a spread-spectrum signal embedded with psychoacoustic masking and error correction. It is designed to survive exactly the kinds of processing this tool performs. Results vary by track — quiet passages, sparse arrangements, and certain musical styles may be harder to process without audible degradation.
- High-frequency trade-off. The watermark occupies the upper frequencies (community analysis suggests the 18–24 kHz region). Any processing in this range risks dulling cymbal shimmer, reverb tails, and "air." The tool tries to minimize this, but some quality loss is inherent — especially in quiet passages where the watermark energy is hardest to mask.
- Detection is not binary. Watermark detectors use a correlation threshold — the tool aims to push the correlation statistic below that threshold, but the threshold itself can change with detector updates. What works today may not work tomorrow.
- Platform re-detection. Even if the watermark is attenuated, platforms like Suno also use content fingerprinting (via partners like Audible Magic) and metadata provenance checks. Removing the watermark alone does not guarantee the track will pass upload screening on the originating platform.
- Forensic traces. Any removal process leaves traces — spectral holes, reconstruction artifacts, or statistical fingerprints. A sophisticated detector can often identify that a file was tampered with even when it cannot read the original watermark.
- Legal considerations. Watermark removal on content you own or have rights to is generally permissible. Removing watermarks from content you do not own may violate anti-circumvention provisions (DMCA §1201, EU InfoSoc Directive Art. 6). This tool does not condone misuse.
Download the latest release from the mp3ai.info download page:
👉 https://www.mp3ai.info/projects/suno-watermark-remover/
| Package | Size | Platform |
|---|---|---|
swr-install.exe |
11 MB | Windows x86-64 |
swr.AppImage |
14 MB | Linux x86-64 |
Run it (Windows):
# Download swr-install.exe and run it — a self-contained installer
.\swr-install.exeThe installer sets up Suno Watermark Remover and adds it to your Start Menu.
Run it (Linux):
chmod +x swr.AppImage
./swr.AppImageNo installation required — the AppImage is fully self-contained. You can also double-click it in your file manager.
Note: the repository intentionally ships without the proprietary signal-processing core. The build below reproduces the freeware binary for personal use; the method itself remains closed-source.
# System deps (Debian/Ubuntu)
sudo apt install -y build-essential pkg-config cmake \
libx11-dev libxkbcommon-dev libwayland-dev libegl1-mesa-dev \
libgl1-mesa-dev libfontconfig-dev libxcb-render0-dev \
libxcb-shape0-dev libxcb-xfixes0-dev
cargo build --release
./target/release/suno_wm_remover# Install Rust: https://rustup.rs
cargo build --release
.\target\release\suno_wm_remover.execargo testCovers the FFT/STFT, spectral features, psychoacoustic models, metadata scanning, a full decode→analyze→detect pipeline, and an encode→decode roundtrip proving the pure-Rust MP3 encoder produces valid output.
This tool is free to use. If you find it useful, consider supporting its continued development — the cat-and-mouse treadmill means the method needs regular updates as Suno revises its watermarking scheme.
Freeware. You may use this software for personal, educational, or research purposes. Commercial use is prohibited without explicit permission from the author. The source code is not available.