STEMperator (REAPER scripts)

AI-powered stem separation for REAPER

STEMperator uses AI (Demucs/HTDemucs) to separate audio into individual stems. This repo focuses on REAPER scripts (via ReaPack). The former plugin/standalone is developed separately as STEMdropper; historical JUCE/VST files live in legacy/.

---

Why STEMperator?

Stay in your DAW. No more exporting, switching to external tools, and re-importing.

STEMperator integrates directly into REAPER, letting you:

• Process time selections - Select a 4-bar section and extract just the drums from that part
• Edit in place - Replace a media item with its separated stems instantly
• Work non-destructively - Original files are preserved, stems are created alongside
• Be surgical - Found a sample with an unwanted vocal bleed? Extract only what you need
• Stay creative - Remix, mashup, or repair audio without leaving your project

Real-World Workflows

| Scenario | Traditional Approach | With STEMperator | |----------|---------------------|------------------| | Remove vocal from a loop | Export → External tool → Re-import → Align | Select item → Karaoke → Done | | Extract drums from reference track | Export → Separate → Import 4 files → Create tracks | Select → All Stems → Auto-routed tracks | | Fix a sample with unwanted bass | Find alternate sample or EQ compromise | Select → Bass Only → Delete original bass | | Create acapella from full mix | External software, manual export/import | Select → Vocals Only → New track | | Isolate guitar solo for analysis | Complex EQ/phase tricks | Select region → Guitar Only |

---

Quick Start (REAPER)

1. Install AI Backend (one time, ~5 minutes):

   Windows (PowerShell as Admin):

   # Download and run installer
   Invoke-WebRequest -Uri "https://raw.githubusercontent.com/flarkflarkflark/Stemperator/main/scripts/install.ps1" -OutFile install.ps1
   .\install.ps1
   

   macOS/Linux (Terminal):

   curl -sSL https://raw.githubusercontent.com/flarkflarkflark/Stemperator/main/scripts/install.sh | bash
   

2. Install REAPER Scripts:

   • Open REAPER → Extensions → ReaPack → Import repositories
   • Paste: https://raw.githubusercontent.com/flarkflarkflark/Stemperator/main/scripts/reaper/index.xml
   • Extensions → ReaPack → Browse packages → Search "STEMperator" → Install

3. Verify Setup:

   • Actions → Run: "STEMperator: Installation & Setup"
   • Should show all green checkmarks ✓

4. Use It:

   • Select audio item → Actions → "STEMperator: AI Stem Separation"

---

Features

Stem Types

| Model | Stems | Best For | |-------|-------|----------| | htdemucs (4-stem) | Vocals, Drums, Bass, Other | Fast, general use | | htdemucs_ft (4-stem) | Vocals, Drums, Bass, Other | Highest quality | | htdemucs_6s (6-stem) | Vocals, Drums, Bass, Other, Guitar, Piano | Detailed separation |

Flexible Input Selection

STEMperator works with whatever you have selected:

| Selection Type | What Gets Processed | |----------------|---------------------| | Media item(s) | Full item(s) - stems replace or appear alongside original | | Time selection | Only the selected time range - perfect for surgical edits | | Time selection + item | Intersection of both - maximum precision |

This means you can:

• Process an entire song by selecting the media item
• Extract stems from just a chorus by making a time selection
• Work on a specific 2-bar section without affecting the rest

REAPER Scripts

| Script | Description | |--------|-------------| | STEMperator: AI Stem Separation | Main dialog - full control over model, stems, and options | | STEMperator: Karaoke | One-click vocal removal (keeps drums, bass, other) | | STEMperator: Vocals Only | Extract vocals to new track | | STEMperator: Drums Only | Extract drums to new track | | STEMperator: Bass Only | Extract bass to new track | | STEMperator: All Stems | Extract all stems to separate tracks | | STEMperator: Explode Stems | Route stems to individual tracks | | STEMperator: Setup Toolbar | Add quick-access toolbar buttons |

Keyboard Shortcuts (in dialog)

• 1-6 - Toggle stems
• K - Karaoke preset
• I - Instrumental preset
• A - All stems
• Enter - Start separation
• Esc - Cancel

---

Installation Details

Requirements

• Python 3.9+ (3.10-3.12 recommended)
• ffmpeg (for audio conversion)
• ~2GB disk space (for AI models)

Automatic Installation

The install scripts automatically:

1. Find or install Python
2. Create isolated virtual environment
3. Install PyTorch with GPU support (if available)
4. Install audio-separator package
5. Pre-download AI models

GPU Acceleration

| GPU | Support | Speed Improvement | |-----|---------|-------------------| | NVIDIA (CUDA) | ✓ Full | 10-20x faster | | AMD (ROCm) | ✓ Linux only | 10-15x faster | | Apple Silicon (MPS) | ✓ Native | 5-10x faster | | Intel/AMD (CPU) | ✓ Fallback | Baseline |

Manual Installation

If automatic installation fails:

Windows:

python -m pip install audio-separator[gpu]
winget install Gyan.FFmpeg

macOS:

pip3 install audio-separator[gpu]
brew install ffmpeg

Linux (Ubuntu/Debian):

pip3 install audio-separator[gpu]
sudo apt install ffmpeg

Linux (Arch):

pip install audio-separator[gpu]
sudo pacman -S ffmpeg

---

Legacy plugin/standalone

The VST3/standalone implementation has moved to the STEMdropper project. Historical JUCE/VST code and resources remain in legacy/ for reference.

---

Processing Time

| Hardware | 3-minute song | |----------|---------------| | RTX 3080 / RX 6800 | 15-30 seconds | | RTX 2070 / RX 5700 | 30-60 seconds | | Apple M1/M2 | 45-90 seconds | | Modern CPU (8 core) | 2-5 minutes | | Older CPU | 5-15 minutes |

---

Output

Stems are saved as WAV files alongside your original audio (regardless of input format):

mysong.mp3           # Your original file (mp3, wav, flac, etc.)
mysong_vocals.wav    # Stems are always high-quality WAV
mysong_drums.wav
mysong_bass.wav
mysong_other.wav
mysong_guitar.wav    # 6-stem only
mysong_piano.wav     # 6-stem only

---

Troubleshooting

"Python not found"

Run the install script for your platform, or install Python 3.10+ manually.

"audio-separator not found"

pip install audio-separator[gpu]

"ffmpeg not found"

• Windows: winget install Gyan.FFmpeg
• macOS: brew install ffmpeg
• Linux: sudo apt install ffmpeg or sudo pacman -S ffmpeg

Slow processing

• Ensure GPU drivers are up to date
• Use htdemucs model (faster than htdemucs_ft)
• Check GPU is being used (look for CUDA/ROCm/MPS in output)

Out of memory

• Close other GPU-intensive applications
• Try htdemucs instead of htdemucs_6s
• Processing in smaller chunks (time selection)

---

Credits

• Demucs: Meta AI Research (MIT License)
• audio-separator: Beverly Nguyen (MIT License)
• JUCE: RAL/ROLI (Commercial/GPL)

---

License

MIT License - see LICENSE

---

flarkAUDIO - Professional Audio Tools