Home
A2M is a Windows application that turns piano recordings into an editable MIDI file. In practical terms, it listens to overlapping pieces of a recording, estimates which piano notes and pedals are active, refines their timing and strength, joins the pieces into one timeline, and writes a .mid file.
Transcription runs locally on your computer. With your permission, A2M may connect to the internet to download its model, optional GPU components, or application updates. Your source recording is not uploaded for transcription.
- Convert a recording: Quick Start
- Learn how A2M works: How A2M Works
- Review a worked example: Worked Example and Validation
- Improve transcription quality: Accuracy, Limitations and Best Results
- Review supported files and outputs: Input and Output Files
- Resolve common issues: Troubleshooting
The documentation covers the full processing path from decoded audio through spectral analysis, model inference, interval decoding, timing refinement, section stitching, and final MIDI construction. Each concept is presented visually before the underlying implementation is explained.
- Audio Processing Pipeline
- A2M Piano Engine
- From Model Output to MIDI
- CPU and GPU Acceleration
- Developer Architecture
These pages describe the current A2M project as a whole: what users give it, how the transcription process works, what it produces, how to interpret the result, where its limits are, and how the implementation is organized. Version-sensitive details are identified where they matter.