Automatic Piano Transcription

Music transcription is the process of creating a musical score (i.e. a symbolic representation) from an audio recording. Although expert musicians are capable of transcribing polyphonic pieces of music, the process is often difficult for complex recordings. We have been pursuing a classification-based approach to this task. This page links to some of our results, as well as some of the ground-truth data we have prepared in this project.

Data

The MIDI data used in our experiments was collected from the Classical Piano MIDI Page. In addition to the synthesized audio, we made piano recordings from a subset of the MIDI files using a Yamaha Disklavier piano. Below are the live training and testing sets for algorithm development.

• Tar file of the training data
• Tar file of the test data

The REF files are saved as MIDI files. The corresponding audio files are in standard .wav format.

Papers

Here are some of our publications that relate to this data:

G. Poliner, D. Ellis (2006). A Discriminative Model for Polyphonic Piano Transcription

to appear, EURASIP Journal on Applied Signal Processing 2006. (16pp)

Acknowledgment

This material is based in part upon work supported by the National Science Foundation under Grants No. IIS-0238301 and IIS-07-13334. Any opinions, findings and conclusions or recomendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the National Science Foundation (NSF).

Last updated: $Date: 2008/06/11 19:21:49 $
Graham Poliner <graham@ee.columbia.edu>