{"id":50787962,"url":"https://github.com/juxstin1/stemsep","last_synced_at":"2026-06-12T09:04:38.810Z","repository":{"id":363940890,"uuid":"1265643212","full_name":"juxstin1/stemsep","owner":"juxstin1","description":"GPU-accelerated stem separation deck for DJs - Tauri + Demucs/BS-Roformer on AMD ROCm for Windows. Drop a track, pull acapellas, drums, bass \u0026 synths as DAW-ready WAVs.","archived":false,"fork":false,"pushed_at":"2026-06-11T01:14:42.000Z","size":391,"stargazers_count":0,"open_issues_count":0,"forks_count":0,"subscribers_count":0,"default_branch":"main","last_synced_at":"2026-06-11T03:10:38.974Z","etag":null,"topics":["amd","audio-separation","demucs","dj-tools","react","rocm","rust","stem-separation","tauri"],"latest_commit_sha":null,"homepage":null,"language":"TypeScript","has_issues":true,"has_wiki":null,"has_pages":null,"mirror_url":null,"source_name":null,"license":null,"status":null,"scm":"git","pull_requests_enabled":true,"icon_url":"https://github.com/juxstin1.png","metadata":{"files":{"readme":"README.md","changelog":null,"contributing":null,"funding":null,"license":null,"code_of_conduct":null,"threat_model":null,"audit":null,"citation":null,"codeowners":null,"security":null,"support":null,"governance":null,"roadmap":null,"authors":null,"dei":null,"publiccode":null,"codemeta":null,"zenodo":null,"notice":null,"maintainers":null,"copyright":null,"agents":null,"dco":null,"cla":null}},"created_at":"2026-06-11T00:48:43.000Z","updated_at":"2026-06-11T01:14:46.000Z","dependencies_parsed_at":null,"dependency_job_id":null,"html_url":"https://github.com/juxstin1/stemsep","commit_stats":null,"previous_names":["juxstin1/stemsep"],"tags_count":null,"template":false,"template_full_name":null,"purl":"pkg:github/juxstin1/stemsep","repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/juxstin1%2Fstemsep","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/juxstin1%2Fstemsep/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/juxstin1%2Fstemsep/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/juxstin1%2Fstemsep/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/juxstin1","download_url":"https://codeload.github.com/juxstin1/stemsep/tar.gz/refs/heads/main","sbom_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/juxstin1%2Fstemsep/sbom","scorecard":null,"host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":286080680,"owners_count":34236583,"icon_url":"https://github.com/github.png","version":null,"created_at":"2022-05-30T11:31:42.601Z","updated_at":"2026-05-26T15:22:16.424Z","status":"online","status_checked_at":"2026-06-12T02:00:06.859Z","response_time":109,"last_error":null,"robots_txt_status":"success","robots_txt_updated_at":"2025-07-24T06:49:26.215Z","robots_txt_url":"https://github.com/robots.txt","online":true,"can_crawl_api":true,"host_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub","repositories_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories","repository_names_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repository_names","owners_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners"}},"keywords":["amd","audio-separation","demucs","dj-tools","react","rocm","rust","stem-separation","tauri"],"created_at":"2026-06-12T09:04:36.802Z","updated_at":"2026-06-12T09:04:38.803Z","avatar_url":"https://github.com/juxstin1.png","language":"TypeScript","funding_links":[],"categories":[],"sub_categories":[],"readme":"\u003cdiv align=\"center\"\u003e\n\n# StemSep\n\n**GPU-accelerated stem separation deck for DJs.**\nDrop a track → pull acapellas, drums, bass \u0026 synths as DAW-ready WAVs.\nBuilt for AMD GPUs on Windows — ROCm-accelerated PyTorch, no CUDA required.\n\n\u003cp\u003e\n  \u003cimg src=\"docs/stemsep-progress.webp\" width=\"56%\" alt=\"StemSep — GPU separation in progress with live engine log\" /\u003e\n  \u003cimg src=\"docs/stemsep-rack.webp\" width=\"43%\" alt=\"StemSep — 4-stem rack with synced waveforms, solo/mute per stem\" /\u003e\n\u003c/p\u003e\n\n\u003c/div\u003e\n\n## Why\n\nUltimate Vocal Remover and friends leave Radeon owners on CPU or flaky DirectML. StemSep runs the same model families — **BS-Roformer** and **Demucs** — on AMD's official ROCm-on-Windows PyTorch wheels. A 4-stem fine-tuned Demucs pass on an RX 9070 XT takes ~75 seconds; the same job on CPU takes 12–20 minutes.\n\n## Modes\n\n| Preset | Model | Stems |\n|---|---|---|\n| Acapella / Instrumental | BS-Roformer ViperX 1297 | Vocals, Instrumental |\n| DJ 4-Stem | Demucs htdemucs_ft | Vocals, Drums, Bass, Other |\n| Producer 6-Stem | Demucs htdemucs_6s | Vocals, Drums, Bass, Guitar, Piano, Other |\n\nSynths/pads land in **Other** on the Demucs presets. The Roformer 2-stem beats Demucs (and UVR's defaults) for clean acapellas.\n\nAfter a run you get a synced multi-waveform rack: space-bar transport, click-to-seek across all stems, per-stem solo/mute — audition the split before you commit it to a deck or a project. Right-click any stem to save just that piece as a WAV wherever you want it, or reveal it in Explorer — or grab a stem's label and **drag it straight out of the app** onto a rekordbox deck, an Ableton track, or an Explorer folder.\n\n## Requirements\n\n- Windows 11, AMD RX 9000 / RX 7900-series GPU (ROCm-on-Windows supported)\n- AMD Adrenalin driver **26.2.2+** (older drivers → falls back to CPU, much slower)\n- [uv](https://docs.astral.sh/uv/) + Python 3.12, Node 20+, Rust toolchain\n\n## Setup\n\n```powershell\nnpm install\nuv sync --project python   # ~2 GB: ROCm PyTorch + audio-separator\nnpm run tauri dev          # dev mode\nnpm run tauri build        # installer\n```\n\nFirst separation per model downloads the model checkpoint (~100–700 MB) into the audio-separator cache.\n\n## Output\n\nStems land next to the source track — drag straight into rekordbox or Ableton:\n\n```\n\u003ctrack folder\u003e\\Stems\\\u003ctrack name\u003e\\\u003cpreset\u003e\\\u003ctrack name\u003e - Vocals.wav\n```\n\n## Architecture\n\n- **Tauri 2 + React 19** UI (`src/`): drop zone, preset cards, synced multi-waveform stem rack (wavesurfer.js), solo/mute, space-bar transport.\n- **Rust** (`src-tauri/`): spawns the Python sidecar via `uv run`, streams JSON-lines progress as Tauri events, kill-on-abort.\n- **Python sidecar** (`python/separate.py`): [audio-separator](https://github.com/nomadkaraoke/python-audio-separator) with ROCm torch pinned by direct wheel URLs in `pyproject.toml` (so a stray `pip install` can't silently swap in CPU torch — a known ROCm-on-Windows footgun). AMD's torch wheel requires `rocm[libraries]`, which only exists as an sdist in AMD's wheel directory — a uv dependency override redirects the requirement there.\n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fjuxstin1%2Fstemsep","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Fjuxstin1%2Fstemsep","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fjuxstin1%2Fstemsep/lists"}