beiciliang / eusipco2018-legatopedal

Companion code for EUSIPCO 2018 paper "Piano Legato-Pedal Onset Detection Based on a Sympathetic Resonance Measure"

eusipco2018-legatopedal

Companion code for the paper:

Beici Liang, György Fazekas, Mark Sandler. "Piano Legato-Pedal Onset Detection Based on a Sympathetic Resonance Measure​", in Proceedings of the 26th European Signal Processing Conference (EUSIPCO), 2018. (accepted)

Index

• get_features.py: Main python script to extract features from the example audio file in input/chopin. Save the features as a .npz file in features.

• evaluate.py: Python script to run the task of aic or logr based on features/chopin_features.npz (extracted features) and input/chopin/chopin_gt.npy (ground truth). If set the task as logr_eusipco, return the performance matrix of logistic regression model using the data in eusipco-data. The matrix corresponds to the table of experiment result in the paper.

• input: The folder contains .csv files of f0 and inharmonicity coefficient computed from 88 piano notes (check this repository for calculation details). The sub-folder input/chopin includes example audio file chopin.wav, ground truth chopin_gt.npy obtained from chopin.mid, and transcription result chopin_transcription.npy (check this repository for transcription details).

Requirements

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

Software	Version
Python	2.7.14 64bit
IPython	5.3.0
OS	Darwin 15.6.0 x86_64 i386 64bit

Other dependencies can be installed by

pip install -r requirements.txt