Model Selection Matrix
Model Profiles
Section titled “Model Profiles”| Dimension | 4-Stem Model (htdemucs) |
6-Stem Model (htdemucs_6s) |
|---|---|---|
| Output Stems | Vocals, Drums, Bass, Other | Vocals, Drums, Bass, Other, Guitar, Piano |
| Target Use Cases | Vocal removal, karaoke, acapella extraction, rhythm sampling | Multitrack remixing, acoustic guitar isolation, piano transcription |
| Recommended RAM | $\ge 3.0\text{ GB}$ available | $\ge 4.0\text{ GB}$ available |
| Peak RSS (CPU) | $\sim 2.2\text{ GB} - 2.6\text{ GB}$ | $\sim 2.8\text{ GB} - 3.4\text{ GB}$ |
| Real-Time Factor (RTF) | $0.35 - 0.55$ | $0.60 - 0.95$ |
| Model Size on Disk | $\sim 80\text{ MB}$ | $\sim 120\text{ MB}$ |
| CLI Flag | --stems 4 |
--stems 6 |
| MCP Parameter | "stems_count": 4 |
"stems_count": 6 |
Model Checkpoint Caching
Section titled “Model Checkpoint Caching”Models are downloaded on first invocation and stored locally:
- Default directory:
~/.cache/audio-tools/models - Download weights explicitly:
uv run audio-tools download-models --model all