Miditizer
A piano transcription model published on Hugging Face. It takes raw audio and produces the notes that were played — a fully automatic music transcription pipeline for piano recordings.
What it does
Given an audio recording, the model outputs the notes the pianist played, each with pitch, timing, and a loudness estimate, written to MIDI. The published pipeline is set up as an audio-classification task, with weights shipped as a PyTorch model.
Where to find it
- Miditizer on Hugging Face — the model card and weights (gated on HF; account and terms acceptance required)
- citan’s Hugging Face profile — hub page that links both models
Technical Details
- Task: audio-classification (piano transcription)
- Framework: PyTorch
- License: CC-BY-NC-SA-4.0
- Reference: arXiv:2404.09466 (third-party comparison system, linked from the model card)