Modelling MIDI data with Machine Learning

At its surface, PianoRoll serves as a platform for pianists to track and share their progress with friends and the broader community. Yet, it’s so much more than that.

Pianoroll logoPianoroll logo text

Audio, images and text already benefit from well-established deep learning architectures and processing pipelines that consistently yield amazing results.

Recording piano performances in MIDI format introduces a new and exciting area of research, combining challenges from text, image and audio into a single modality. MIDI format services as a bridge, capturing numerous nuances of piano performances in a structured data format.

PianoRoll

PianoRoll is a platform for tracking effort and hosting piano competitions. PianoRoll users provide crowd-sourced, high-quality MIDI recordings rich with performance nuances. This allows us to curate datasets for cutting-edge machine learning research.

PianoRoll users provide crowd-sourced, high-quality MIDI recordings rich with performance nuances. This allows us to curate datasets for cutting-edge machine learning research.

Python Tools for MIDI Processing

Fortepyan python library
Fortepyan python library
  • MIDI - pandas interface
PianoRoll streamlit widget
PianoRoll streamlit widget
  • Visualisation and playback
Huggingface dataset
Huggingface dataset
  • 500 hours
  • 10000000 tokens
  • Ethical sourcing

Pydata Talk

We presented at PyData London 2024. Check out our speech on modeling the emotional nuances of music performances.
Lets open the door to a mathematical understanding of music together.

Stay up to date with PianoRoll

Stay up to date with Piano for AI

Looking for Data Science Services?

EPR Labs logo white
Make Data Science Great Again