# AudioGhost AI

AudioGhost AI is a local, GPU-powered audio separation tool that uses Meta's SAM-Audio model to extract or remove any sound from a track using a natural-language description.

Canonical page: https://ourlocalhost.com/app/7765/audioghost-ai

## Description

AudioGhost AI is a local, GPU-powered audio separation tool that uses Meta's SAM-Audio model to extract or remove any sound from a track using a natural-language description.

## Platform availability

Available on 1 of 10 platforms: Linux.

## Desktop and other devices

Linux.

## Health score

59/100 (Fair), based on 4 of 5 factors.
- Maintained: 72 (last commit 5 days ago)
- Popular: 56 (606 GitHub stars)
- Easy to install: 25 (1 install platform, install notes)
- Security check: 72 (no serious findings, 3 minor, tests)

## Links

- GitHub: https://github.com/0x0funky/audioghost-ai

## Common questions

### Does v1.1 reduce the first-run download size?

Yes. v1.1 lite mode skips ImageBind, CLAP, SAM-Audio-Judge, and PE-A-Frame, cutting the initial download by about 25 GB.

### Does the v1.1 installer prevent Windows setups from using a CPU-only PyTorch build?

Yes, v1.1 pins torch and xformers so pip cannot swap in a CPU build. It uses CUDA 12.8 for RTX 50-series GPUs.

### How are long audio files processed?

Long files are processed in overlapping chunks with crossfades. This avoids seams and missing audio at the end.

### What output files can AudioGhost produce after a separation?

It can export 24-bit WAV, FLAC, MP3, or a ZIP with readable names. For video, the separated audio can be returned with the video.

### Can AudioGhost be used without the graphical UI?

Yes, there is a CLI entry point: python sam_audio_lite.py song.mp3 "vocals".

---
Source: Our Local Host (https://ourlocalhost.com/app/7765/audioghost-ai), the self-hosted and open-source app directory.
