SheetSage2 is a lightweight AI music foundation model that converts raw audio recordings (.mp3, .wav) directly into editable lead sheets, multi-track MIDI files, and ABC musical notation.
| Step 1: Input | Step 2: AI Action | Step 3: Result |
|---|---|---|
| Raw Audio Track Load any MP3 or WAV recording |
SheetSage2 Model Extract pitch, beat grid & chords |
Score & MIDI Export Get editable ABC score & MIDI file |
- 📦 Tiny Model Footprint: Only 229 MB in size, running smoothly on CPU or local GPUs with under 4GB VRAM.
- 🎼 Multi-Format Output: Generates printable lead sheets (
score.abc), multi-track MIDI (transcription.mid), and timed JSON event annotations. - 🏆 Top Benchmark Accuracy: Achieves 92.9% downbeat precision and 90.08% chord recognition score.
- 🚀 Hugging Face Native: Built on MERT-v2 FullSong acoustic features with 3-line Python pipeline execution.
Copy and paste this single PowerShell command to install all dependencies and download the model checkpoint directly:
pip install torch torchaudio transformers soundfile mir_eval mido pretty_midi huggingface_hub; hf download m-a-p/SheetSage2 --local-dir models/SheetSage2import torch
from transformers import AutoModel
# Load SheetSage2 model locally or from Hugging Face
model = AutoModel.from_pretrained(
"m-a-p/SheetSage2", trust_remote_code=True
).eval().to("cuda" if torch.cuda.is_available() else "cpu")
# Transcribe audio file into MIDI and ABC score
result = model.transcribe("inputs/piano_sample.wav", output_dir="outputs")- 🎸 Song Transcriptions for Musicians: Instantly convert recorded acoustic jams into printed sheet music.
- 🎹 DAW Remixing & Beat Making: Import extracted multi-track MIDI files directly into FL Studio or Ableton.
- 🎤 AI Vocal Cover Creation: Extract clean lead sheets to feed melody lines into YuE2 music generators.
- 📚 Music Education & Analysis: Analyze song chord progressions, key signatures, and measure structures.
- 🔍 Audio Archiving & Cataloging: Generate searchable text annotations (
events.json) for large audio libraries.
- 🎛️ Live Real-Time Audio Transcription: Stream microphone audio directly into live scrolling sheet music.
- 🎻 Orchestral Instrument Separation: Expanded multi-stave separation for complex classical ensemble scores.
- 📱 Mobile & Edge Export: Quantized ONNX weights for fast execution on tablets and mobile devices.
- 🎼 Guitar Tablature Converter: Automatic conversion from ABC notation to 6-string guitar tabs.
- 🌐 WebGPU Web Interface: Browser-based drag-and-drop transcription tool powered by local WebGPU.
SheetSage2/
├── main.py # Main production transcription pipeline script
└── README.md # Project documentation and guide
SheetSage2 AI Music Transcription Audio to MIDI Lead Sheet Generator MERT-v2 ABC Notation Python Music AI Sheet Music Transcriber AI Chord Recognition Local AI Model