A demonstration is available in the Demo.ipynb jupyter notebook.
- rnn_conv.ipynb is a notebook with the model, it's training and it's testing.
- rnn_conv_training.py is a script to train the model.
- build_dataset.py is a script to build the tfRecords dataset from the audios of the LibriSpeech Corpus
- Other files are tools for the first ones
Some files are taken from the Stephenson et al. paper (https://github.com/lab41/magnolia). These are the files classifying LibriSpeech Corpus into male and female files in a computer friendly way (the classification comes from the LibriSpeech corpus) Note that there was a mistake in the classification that was corrected (in the dev set)
The computation of the indices was taken from https://github.com/craffel/mir_eval.