magenta/mt3

MT3: Multi-Task Multitrack Music Transcription

What it solves

It provides a way to automatically transcribe audio recordings of music into musical notation, supporting both single-instrument (piano) and multi-instrument recordings.

How it works

MT3 uses the T5X framework to implement a multi-task multitrack music transcription model. It allows users to apply pretrained checkpoints—either a piano-specific model or a multi-instrument model—to convert audio files into transcription data.

Who it’s for

Musicians, researchers, and developers who need to convert audio recordings of music into digital notation.

Highlights

  • Multi-instrument support: Capable of handling multiple instruments simultaneously.
  • Pretrained checkpoints: Includes models based on on research papers from ISMIR 2021 and ICLR 2022.
  • Colab integration: Provides a notebook for easy testing and transcription of personal audio files.

Related

  • Project
  • Project
  • Project
  • Project