ICASSP18

Code to reproduce the experiments for the paper:

@inproceedings{kelz_icassp18,
  author    = {Rainer Kelz and Gerhard Widmer},
  title     = {Investigating Label Noise Sensitivity Of Convolutional Neural Networks For Fine Grained Audio Signal Labelling}
  booktitle = {2018 {IEEE} International Conference on Acoustics, Speech and Signal
               Processing, {ICASSP} 2018, Calgary, Alberta, Canada, April 15-20, 2018},
  year      = {2018}
}

Installation Instructions (some assembly required)

All of the following assumes that you already properly set up your CUDA environment, and have working numpy, scipy, theano, lasagne installations.

• obtain the MAPS dataset (http://www.tsi.telecom-paristech.fr/aao/en/2010/07/08/maps-database-a-piano-database-for-multipitch-estimation-and-automatic-transcription-of-music/)

• clone this repository

• edit setup.sh to let it know about the folder you extracted the dataset to

• execute setup.sh, it'll create some directories, and symlinks

• execute python setup.py develop, to install the package and all needed dependencies

• go to the bin directory and execute python run.py train maps_labelnoise_conv

• if everything went well, you'll see a printout of the exact net definitions, parameter count and a progress bar. This will be very slow in the beginning, as it needs to decode audio data / compute all (spectrogram, label) pairs. (I removed the joblib cache again, b/c there were some weird troubles using it. You may of course put it back into datasources/maps_labelnoise.py.)

• after training, go to bin and execute python run.py test maps_labelnoise_conv

• all results should be in the folders bin/runs/maps_labelnoise_conv/<run-id> having names native_fps_result.pkl and high_fps_result.pkl