sustain-pedal-detection

Companion codes for the submission:

Beici Liang, György Fazekas, Mark Sandler. "Piano Sustain-Pedal Detection Using Convolutional Neural Networks".

Index

• 0. pedal midi info.ipynb: understand MIDI files and how the ground-truth annotations are extracted

• 1. dataset preparation.ipynb: how to build the dataset and generate excerpts

• 2.1 pedal onset classification.ipynb: how to train Conv2D-onset

• 2.2 pedal segment classification.ipynb: how to train Conv2D-segment

• 2.3 how mfcc performs on the small dataset.ipynb: compare with SVM using MFCC features

• 3. piece-wise detection.ipynb: how to fuse the decision outputs from Conv2D-onset and Conv2D-segment so as to perform the detection on a piano piece.

Trained models are saved in folder ./save-model. Evaluation results for every piece in the testing set are saved in psegment-testresult_onset98_seg98.csv.

Box plot of F1 score and bar plot of pedal-frame proportion ordered by composer's lifetime:

Requirements

Codes are based on the following settings and their corresponding versions.

Setting	Version
OS	Centos 7.3
GPU	Titan Xp
module	cuda/8.0-cudnn5.1
Python	2.7.5

Python dependencies can be installed by

pip install -r requirements.txt

You need to install Jupyter Notebook to run .ipynb in your local browser.