# SoundScrub

> SoundScrub is a desktop application that uses Meta's SAM Audio (Segment Anything Model for Audio) to target specific sounds in video files. Users describe a sound in natural language, and the AI separates it from the rest of the audio.

## Three Output Modes

- **Remove**: Completely remove the targeted sound from the video.
- **Keep**: Keep only the targeted sound and strip everything else.
- **Adjust**: Control the volume of the targeted sound and everything else independently with sliders, then mix them back together locally.

## How It Works

1. User selects a video file (stays on their machine, never uploaded).
2. Audio is extracted and sent to the cloud for AI processing.
3. The AI separates the audio into two streams: the targeted sound and everything else.
4. For remove/keep mode, the app replaces the video's audio with the appropriate stream.
5. For adjust mode, the app downloads both streams and lets the user control the volume of each with sliders before mixing locally.

## Supported Sounds

Speech, background music, ambient noise, animal sounds, urban noise, or any custom sound described in plain text (e.g. "dog barking", "keyboard clicking", "wind noise").

## Technical Details

- Powered by Meta's SAM Audio model (Segment Anything Model for Audio)
- Desktop app built with Electron (macOS, Windows, Linux)
- Audio processing happens in the cloud on GPU infrastructure
- Video is never uploaded - only extracted audio leaves the user's machine
- Volume mixing in adjust mode uses ffmpeg locally
- Original video and audio quality is preserved

## Pricing

$0.25 per 30 seconds of video. Credits purchased in advance, never expire. No subscriptions.

## Links

- Website: https://soundscrub.video
- About: https://soundscrub.video/about
- Pricing: https://soundscrub.video/pricing
- API (coming soon): https://soundscrub.video/sam-audio-api
- Privacy: https://soundscrub.video/privacy
- Terms: https://soundscrub.video/terms
- Refund Policy: https://soundscrub.video/refund
- Support: support@soundscrub.video