Keep the shot.
Clean the sound.

Clean distracting audio in Seedance, Genjutsu and other generations, or work with existing MP4, MP3 and WAV files. Drop one file, choose what stays, compare the layers and download audio for your edit.

Download Windows EXE / Download Apple Silicon DMG / Download Intel Mac DMG

Made by cyr4x. Made for the Higgsfield Community.
Version 1.3.0 / Documentation refreshed September 26, 2026.
Community-created; not an official Higgsfield product.

Install on Windows.

Windows 10/11 x64 / Electron desktop app. Python is bundled. Engine and model downloads happen inside the app on first use.

1. Download the EXE

   Download Windows EXE from the official release. Read the Windows warning page and verify the downloaded file.

2. Quit old copies, then install

   Quit an older app or use Close SoundShredder on its standalone setup page. Run the EXE, choose the app installation folder, and finish. No separate Python, PATH edits or administrator rights are needed.

3. Open and choose storage

   Open SoundShredder from Start or its desktop shortcut. Use Choose folder... for engines, models and sessions, or keep the current location. This storage choice is separate from the app installation folder.

4. Set up the engine

   Select CPU or a compatible NVIDIA GPU, then Set up SoundShredder. Keep internet connected. Allow at least 4 GiB free for CPU or 14 GiB for NVIDIA, plus the app, models and media. NVIDIA needs a compatible driver; no separate CUDA Toolkit is needed.

5. Use it, quit and reopen

   The workspace opens automatically. Finish or cancel processing before quitting. Reopen the same shortcut next time; compatible engines and saved sessions are reused.

Uninstall: quit, then use Uninstall SoundShredder in Start or Windows Settings > Apps. Sessions and engines are retained. Full Windows guide and removal instructions.

Understand the warning.

Our official download source is github.com/xD4O/SoundShredder/releases/latest. The Windows installer is still unsigned. The signed Mac release does not sign the Windows EXE.

1. Confirm the exact source

   Use the xD4O/SoundShredder repository and its release Assets. Avoid reuploaded installers and lookalike projects. Open the official release.

2. Match the checksum

   Download the EXE and SHA256SUMS.txt from the same release. Run Get-FileHash -Algorithm SHA256 -LiteralPath followed by the quoted path to your downloaded EXE. Compare the entire hash against its filename in the checksum file. Windows download verification and warning guide includes a copyable command.

3. Keep protections enabled

   A matching hash checks the bytes against the published file; it is not a malware scan. A named threat detection or a mismatched hash needs investigation. Do not run a mismatched download or disable security protection.

4. Choose whether to continue

   Only after verifying and trusting the download, Edge may offer Keep > Show more > Keep anyway; Windows may offer More info > Run anyway. Availability varies. If blocked without an override, use the source version where permitted or contact your administrator.

Microsoft: how SmartScreen reputation works / Windows signing roadmap

Install on your Mac.

macOS 13 or newer / Electron desktop app. Current 1.3.0 Apple Silicon and Intel builds are Developer ID signed and Apple-notarized. Mac audio processing uses CPU; Metal/MPS is not enabled.

1. Quit any older SoundShredder copy

   Use its Quit command or, for older browser standalones, Close SoundShredder on the setup page. Keep your saved engine and session folders.

2. Move the app into Applications

   Open the matching DMG, drag SoundShredder into Applications, replace the old app if needed, then eject the disk image. Open SoundShredder from Applications.

3. Choose storage and set up

   Optionally use Choose folder..., then Set up SoundShredder. Allow at least 3 GiB free, plus the app, models and saved media. Internet is required initially. No Homebrew, separate Python or developer membership is needed to use the app.

4. Quit and return

   Finish or cancel audio work before quitting. During setup, Cancel setup and quit is available. Reopen from Applications or the Dock; completed sessions remain saved.

Uninstall: quit and move SoundShredder.app to Trash. Saved data remains in the chosen folder and the user profile. Mac storage, removal and troubleshooting.

Prefer your browser?

The Python source version runs on your own computer at localhost. It uses the same workspace. Choose this route if you want source access; Electron is the easier installation for most users.

Download the source ZIP. Extract the entire folder into a writable location such as Documents. Keep all files together.

1. Wait for initial setup

   The launcher downloads the CPU or selected NVIDIA engine. First processing also downloads model weights. The browser normally opens http://127.0.0.1:7860.

2. Keep the launcher window open

   Closing the browser tab does not stop the server. Use Ctrl+C on Windows or Control+C on Mac to stop it. Reopen the same launcher next time.

3. Repair Mac source certificates if needed

   For Python 3.11.9 with CERTIFICATE_VERIFY_FAILED, run Applications > Python 3.11 > Install Certificates.command, then restart SoundShredder. Keep SSL verification enabled. This repairs the source Python environment, not Electron's bundled runtime.

Complete source installation instructions / Mac source SSL troubleshooting and repair launcher

Try it with Akira.

This worked example uses the supplied TestFootage.mp4: approximately 43 seconds, with 48 kHz stereo audio. The screenshots show a completed local Windows source-app run; Electron shares these controls.

1. Drop the MP4 and remove music

   Choose Remove music: Dialogue 100%, Music 0%, Sound effects 100%. Choose Auto, CPU or NVIDIA GPU, then Separate audio. The example used Auto on an RTX 5090; Mac users choose CPU.

2. Compare with the footage

   Enable Show video preview. Switch Listen to between Original and Cleaned mix at the same moment. The monitor uses muted video synchronized to the selected audio.

3. Refine the result

   Audition wanted effects and any voices. If needed, adjust layer levels and click Update mix. Disable Show video preview to return to an audio-only workspace.

Every layer, its own track.

1. Compare one track at a time

   Keep Keep playback position when switching tracks enabled to audition the same moment. Starting a track pauses the others. Preview volume/mute does not change the export.

2. Download the layer you need

   Use the Download control on Dialogue, Music, Sound effects or Removed sounds. Normal separation prepares them automatically.

3. Preparing tracks after Bubble FX

   Use Prepare isolated tracks to hear dialogue/music/effects from the original audio before bubble cleanup. This extra separation preserves the cleaned bubble result; download its added layers individually.

Choose what stays.

Layer presets set the estimated dialogue, music and effects levels. 0% excludes a layer; 100% keeps it. Intermediate levels reduce it. Change levels after separation and click Update mix to reuse the existing stems.

Custom text prompts: Bubble FX currently uses four preset sound descriptions. Use Target a Sound - Guided for a free-form description, preview and saved cleanup version.

Target stubborn bubbles.

Use Bubble FX for a distracting bubble, pop or gurgle while trying to retain other effects. This uses AudioSep separately from the dialogue/music/effects model.

1. Match the unwanted sound

   Choose Water bubbles, Bubbling & popping, Liquid gurgle or Cartoon boing. Start near 85% reduction and restrict the time range when possible.

2. Try aggressive multi-pass for Water bubbles

   Enable Aggressive multi-pass and start with 2 passes. Each pass rechecks the previous cleaned audio. Try 3 or 4 only if stubborn remnants remain. More passes take longer and can affect similar wanted effects.

3. Check the combined removed sound

   Compare Original, Cleaned mix and Removed sounds. The latter combines removal across all passes. Reduce strength, narrow the interval or use fewer passes if wanted effects are caught.

4. Rerun when settings need fresh inference

   Changing sound type, pass count or multi-pass strength/range needs Clean up again, which starts from the saved original in a new session. Single-pass strength/range changes can use Update mix.

Describe the distraction.

A generation has useful voices and impacts, but one repeated sound distracts. Use Target a Sound - Guided to describe that sound instead of removing the whole Effects layer.

1. Prepare your source

   Drop the video or audio file into the separator. Switch to Target a Sound, then Prepare source. The original upload is retained.

2. Describe one sound

   Try a short phrase such as water bubbling, footsteps or a dog barking. The prompt describes what to remove. It is not a guarantee that the model can isolate that sound in every recording.

3. Choose the interval and preview

   Select where the distraction occurs, then create a short preview. Compare Before, Cleaned and Removed. If voices or wanted effects appear in Removed, narrow the interval, lower strength or rephrase.

4. Set strength and passes

   Strength controls how much of the estimated target is subtracted. Passes repeat cleanup on the previous pass. Start with one; use 2-4 for stubborn remnants after checking the sound.

5. Save or continue in steps

   Save a full cleanup version for export. To address another distraction afterward, explicitly use the cleaned version as the next input. Earlier versions and the original remain available.

Full targeted-cleanup guide

Meet your Mixing Lab.

Use the Lab when you want to keep most of a scene and adjust individual moments. The reference picture and all playing stems share one playhead.

1. Upload, then extract

   Choose Mixing Lab, drop a video/audio file, choose CPU/Auto/NVIDIA where available, then click Extract stems. Uploading alone does not extract them. Progress, Cancel and Retry are in this area.

2. Shape one moment

   Choose one second, one nominal frame or a range. Adjust Dialogue, Music or Effects at the playhead. The saved level returns outside the selected interval. 0 dB means unchanged gain; the moving meter shows signal level in dBFS.

3. Listen with the picture

   Use the shared Play control, waveform or scrubber. Solo isolates a channel for listening; it does not change the exported mix. Toggle Video to hide the reference picture.

Clean a stem. Keep the scene.

If bubbles sit in Effects while dialogue and music already sound right, target the Effects stem in the Lab. This keeps the other selected channels out of that cleanup operation.

1. Choose a stem and interval

   Select the stem and time range. Describe the unwanted sound, set strength and 1-4 passes, then preview. Compare Before, Cleaned and Removed before applying.

2. Keep versions available

   Applying cleanup creates a new version for that channel. Removed audio is available to audition or download and does not automatically enter the master mix.

3. Import a replacement

   Bring in an external stem or ambience track when useful. Confirm the import fitting option if its length or format differs; keep the original source as a reference.

4. Export the result

   Export the automated channel WAVs, master WAV, edit report or bundle ZIP. For supported video sources, export an MP4 with the mixed soundtrack. Review the result in motion and check sync in your editor.

Complete Mixing Lab guide

Take it into your edit.

Download a ready-to-use mix or separate layers for more control. Electron uses a native Save dialog; the source web app uses your browser's download behavior.

1. Align the WAV to the original

   Import it into Premiere Pro, DaVinci Resolve or your editor. Place it at the original clip start and mute the original soundtrack. Check sync and transitions.

2. Keep useful isolated layers

   Put music, voices and effects on their own editor tracks. Build your score or foley around the material you want to retain.

3. Save before deleting a session

   Download the outputs to your project folder and keep the processing report. Deleting a SoundShredder session removes its local generated files.

The main separator exports audio. Mixing Lab can also export an MP4 with the mixed soundtrack. Layer separation exports 24-bit PCM WAV; Bubble FX exports 32-bit float WAV. Original media is preserved.

The left panel, explained.

1. Finish or cancel active work

   Wait for processing or click Cancel before quitting. During desktop setup, choose Cancel setup and quit if you need to stop.

2. Quit Electron and reopen normally

   Close its window or use SoundShredder > Quit SoundShredder. Use the Start shortcut or Applications/Dock next time. Opening twice brings the existing window forward.

3. Stop source and older browser launchers correctly

   Closing a browser tab leaves the server running. Stop the source launcher with Ctrl+C / Control+C; older standalone apps have a Close SoundShredder setup control.

Storage, updates and help.

Models: first use downloads about 426 MB for Bandit layer separation or 1.2 GB for Bubble FX. Cached runs can work offline. CPU processing may take longer than the clip duration.

Need help? Include the exact release filename, OS/chip, processing mode and error. Use Setup details or Open logs folder; review private paths before sharing. A named malware detection should be reported separately from an uncommon-download warning.

Report an issue / All documentation / Verification and limits

Windows guide / Mac guide / Mac source SSL repair