Repository navigation
Target a sound. Mix the scene.
SoundShredder 1.3.0 brings experimental text-prompt cleanup, a video-connected Mixing Lab, and in-app desktop updates to Windows and Mac.
Download and install
| Computer | Installer | Instructions |
|---|---|---|
| Windows 10/11, x64 | Windows EXE | Windows installation guide |
| Mac with Apple Silicon | Apple Silicon DMG | macOS installation guide |
| Mac with Intel | Intel DMG | macOS installation guide |
The Mac apps are Developer ID signed and Apple-notarized. Install in Applications. Windows publisher signing remains pending: Edge/SmartScreen may show a reputation or unknown-publisher warning. Read the Windows download guidance and compare your download with SHA256SUMS.txt; keep security protection enabled.
Python is bundled. First setup downloads the audio engine and models. Windows supports CPU or compatible NVIDIA GPUs; this Mac release uses CPU processing. You can choose the engine/model/session storage folder separately from the application location.
What's new
- Target a Sound: describe one unwanted sound, select its interval, preview, and compare Original, Cleaned and Removed. Strength and 1–4 passes are independent. Save the result as a new version while retaining the original. Quick Cleanup keeps the familiar presets and Bubble FX controls.
- Mixing Lab: extract Dialogue, Music and Effects beside your reference video; optionally import Ambience or replacement tracks. Use synchronized playback, meters, mute/solo and time-specific dB edits. Levels return to their saved value outside each edited interval.
- Clean individual stems: preview prompt cleanup on a selected channel, listen to the removed sound and export the master, individual WAVs, stem ZIP or video with the mixed soundtrack.
- Persistent session closing: Close session moves a Lab session into Closed sessions, where it can be reopened with its saved work. Split deeper remains paused.
- Desktop updates: automatic stable-release checks, release notes and download size, background download with progress/cancel, and explicit Restart & update. Later and normal Quit never install an update. Pending saves, uploads, downloads, setup, processing and exports block restart.
Already on 1.2.2 or an older standalone? Install 1.3.0 manually once. This installs the updater for future compatible releases. Existing sessions, storage choices and compatible engines/models are retained. The source/browser edition still updates manually.
Prompt cleanup is experimental: more passes can also remove wanted detail. Preview and listen to Removed before applying. Ambience separation, instrument stems, EQ/compression and recursive splits are not included.
Guides and source edition
- Illustrated community guide — PDF · HTML download
- Updater guide · Mixing Lab
- Optional browser/source packages: Windows/universal ZIP · Mac source ZIP. These require Python setup; most users should choose the Electron installer above.
Verification
Native Windows x64, Apple Silicon and Intel CI exercises installation, real CPU separation, two-pass prompt cleanup, Lab playback/export, storage recovery, close/reopen and full application update/relaunch with retained media. Signed Mac applications must pass Gatekeeper and stapling checks before and after processing and upgrading. See the versioned verification record for evidence and limits.
The first updater baseline uses this candidate's code with a lower test-only version because 1.2.2 has no updater. Functional synthetic-fixture checks do not establish perceptual quality, every consumer GPU/OS combination, or recovery from every power-loss stage. All release assets are listed in SHA256SUMS.txt.
Made by cyr4x, for the Higgsfield Community.