Training General-Purpose Audio Tagging Networks with Noisy Labels and Iterative Self-Verification
For a detailed description of the entire audio tagging system please visit the corresponding github page. In this README I just provide the technical instructions to set up the project.
Before we can start working with the code, we first need to set up a few things:
Setup and Requirements
For a list of required python packages see the requirements.txt or just install them all at once using pip.
pip install -r requirements.txt
To install the project in develop mode run
python setup.py develop --user
in the root folder of the package.
This is what I recommend, especially if you want to try out new ideas.
Getting the Data
Then download the challenge data and organize it in the following folder structure:
<DATA_ROOT> - audio_train - audio_test - train.csv - test_post_competition.csv
Set Data and Model path
In config/settings.py you have to set the following two paths:
DATA_ROOT = "/home/matthias/shared/datasets/dcase2018_task2_release" EXP_ROOT = "/home/matthias/experiments/dcase_task2/"
DATA_ROOT is the <DATA_ROOT> path from above.
EXP_ROOT is where the model parameters and logs will be stored.
Once this is all set up, you can switch to the detailed writeup on this github page.