helenacuesta/continuousf0eval

Continuous Metrics for Evaluating Single-f0 Estimation

continuousf0eval

Companion code for the paper:

"Generalized Metrics for Single-F0 Estimation Evaluation"
Rachel M. Bittner and Juan José Bosch
in International Society for Music Information Retrieval (ISMIR) Conference, 2019

@inproceedings{
  bittner_bosch_2019,
  title={Generalized Metrics for Single-F0 Estimation Evaluation},
  author={Bittner, Rachel M and Bosch, Juan J.},
  booktitle={International Society for Music Information Retrieval (ISMIR) Conference},
  year={2019}
}

Contents

algorithm_outputs

The folder algorithm_outputs contains outputs of the following algorithms:

• crepe
• deep salience
• melodia
• pyin

on the datasets:

• medleydb-pitch
• meldeydb-melody
• ikala
• orchset

experiments

The folder experiments contains the code we used to run the experiments.

• confidence : confidence estimates. confidence/separation contains the confidence computed on the ikala source separated vocals. confidence/stems contains the confidence computed on the clean ikala vocals and on the medleydb-pitch sources.

• compute_confidence.py : the script that was used to generate the confidence files in confidence

• confidence.py : a module for loading the confidence values for different algorithms & ground truth datasets.

• Experiment Plots.ipynb : notebook used to generate Figures 2-6 in the paper

• metrics.py : a module with the proposed metrics implemented

• outputs.py : a module for loading algorithm output files

• plot.py : a module with a few plotting utility functions

• Plots toy examples.ipynb : notebook used to generate Figure 1 in the paper

paper-figs

Generated paper figures.