APNet

This repository includes the code for running the experiments reported in

Zinemanas, P.; Rocamora, M.; Miron, M.; Font, F.; Serra, X. An Interpretable Deep Learning Model for Automatic Sound Classification. Electronics 2021, 10, 850. https://doi.org/10.3390/electronics10070850

Installation instructions

APNet uses DCASE-models and therefore please follow the recomendations from this library:

We recommend to install DCASE-models in a dedicated virtual environment. For instance, using anaconda:

conda create -n apnet python=3.6
conda activate apnet

For GPU support:

conda install cudatoolkit cudnn

DCASE-models uses SoX for functions related to the datasets. You can install it in your conda environment by:

conda install -c conda-forge sox

Before installing the library, you must install only one of the Tensorflow variants: CPU-only or GPU.

pip install "tensorflow<1.14" # for CPU-only version
pip install "tensorflow-gpu<1.14" # for GPU version

Now please install DCASE-models:

pip install "DCASE-models==0.2.0-rc0"

Install other dependencies:

pip install "mirdata>=0.3.0"

Now you can clone and use APNet:

git clone https://github.com/pzinemanas/APNet.git
cd APNet

Usage

Download datasets

cd experiments
python download_datasets.py -d UrbanSound8k

Train models

python train.py -m APNet -d UrbanSound8k -f MelSpectrogram -fold fold1
# Repeat for the other folds

Evaluate models

python evaluate.py -m APNet -d UrbanSound8k -f MelSpectrogram