2024:Polyphonic Transcription

From MIREX Wiki
Revision as of 09:47, 25 August 2024 by Zizzi wang (talk | contribs)
(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)

Task Description

This is our first time introducing polyphonic transcription as a MIREX task. The focus this year is on piano transcription, which involves converting audio piano recordings into symbolic representations like piano rolls, MIDI files, or musical scores. This task is a standard transcription challenge, similar to those implementations in previous studies (Hawthorne et al., 2018; Kong et al., 2020).

Dataset

The model should be trained on MAPS and MAESTRO dataset. We will use the test set of MAPS and MAESTRO for evaluation. Which version?

Metric

We use standard objective metrics:

  • Frame-level metrics
  • Note-level metrics (with or without offset)

These are from mir_eval.

Submission

TBD.

Bibliography