AutoTab

AI-powered guitar tab transcription from MP3 files. Automatically converts audio to readable guitar tabs using state-of-the-art open-source models.

Pipeline

1. Download (optional): Fetch audio from URL using yt-dlp
2. Isolate: Separate guitar stem from other instruments using Meta's Demucs v4 (htdemucs_6s model)
3. Transcribe: Convert isolated guitar to MIDI using Spotify's Basic Pitch
4. Optimize: Transform MIDI into playable guitar tab using Tuttut (prioritizes minimal fretboard distance)

Installation

Prerequisites

• Python 3.8+
• pip

Setup

# Upgrade pip first (important for Python 3.12)
python3 -m pip install --upgrade pip

# Create virtual environment (recommended)
python3 -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate

# Install dependencies
pip install -r requirements.txt

Note: The first time you run Demucs, it will automatically download the model (~500MB). Basic Pitch also downloads models on first use.

Usage

Basic Examples

# Transcribe a local MP3 file
python autotab.py --input path/to/song.mp3

# Transcribe with Drop D tuning
python autotab.py --input song.mp3 --tuning drop_d

# Download and transcribe from YouTube
python autotab.py --url "https://youtube.com/watch?v=..." --tuning standard

Supported Tunings

Tuttut supports various guitar tunings. Common options:

• standard - Standard E tuning (E-A-D-G-B-e)
• drop_d - Drop D tuning (D-A-D-G-B-e)
• d_standard - D Standard (D-G-C-F-A-d)
• half_step_down - Half step down (Eb-Ab-Db-Gb-Bb-eb)

Check Tuttut documentation for full list of supported tunings.

Output

• Intermediate files: Stored in temp/ directory (guitar stems, MIDI files)
• Final tab: Saved to output/ as a .txt file

Example output format:

e|----0-2-3---|
B|----1-3-5---|
G|------------|
D|------------|
A|------------|
E|------------|

Dependencies

• demucs (>=4.0.0) - Audio source separation
• basic-pitch (>=0.2.6) - Music transcription
• tuttut (>=0.1.0) - MIDI to guitar tab conversion
• yt-dlp (>=2023.7.6) - Audio downloading

Troubleshooting

Python 3.12 / numpy installation errors

If you're using Python 3.12 and encounter numpy build errors, ensure you have the latest pip:

python3 -m pip install --upgrade pip

The requirements.txt explicitly includes numpy>=1.24.0 which has pre-built wheels for Python 3.12.

Demucs model not found

Demucs will auto-download the htdemucs_6s model on first run. If it fails:

demucs --repo ~/.cache/demucs -n htdemucs_6s -d cpu your_file.mp3

Basic Pitch errors

Ensure the guitar stem is a clean WAV file. Demucs outputs WAV format by default.

Tuttut tuning options

Use lowercase with underscores: drop_d, d_standard, etc. See tuttut --help for all options.

Project Structure

guitar-tabs/
├── autotab.py          # Main script
├── requirements.txt    # Python dependencies
├── temp/              # Intermediate files (auto-created)
├── output/            # Final tabs (auto-created)
└── README.md          # This file

Limitations

• Transcription accuracy depends on audio quality and guitar complexity
• Demucs v4 isolates 6 stems; guitar stem may contain some bleed from other instruments
• Basic Pitch works best with monophonic (single-note) guitar lines; chords may be less accurate
• Processing can be slow on CPU; GPU acceleration requires additional setup

License

This tool uses open-source AI models. Refer to each model's license:

• Demucs: MIT
• Basic Pitch: CC BY-NC 4.0
• Tuttut: MIT

Contributing

Issues and pull requests welcome for improving transcription quality, adding features, or fixing bugs.