Emergence of music detectors in a deep neural network trained for natural sound recognition

Authors: Gwangsu Kim¹, Dong-Kyum Kim¹, and Hawoong Jeong^1,2

¹ _{Department of Physics, KAIST} ² _{Center for Complex Systems, KAIST}

Introduction

This repo contains source code for the runs in Emergence of music detectors in a deep neural network trained for natural sound recognition

Installation

Supported platforms: MacOS and Ubuntu, Python 3.7

Installation using Miniconda:

git clone https://github.com/kgspiano/Music.git
cd Music
conda create -y --name music python=3.7
conda activate music
pip install -r requirements.txt
python -m ipykernel install --name music

To enable gpu usage, install gpu version torch package from PyTorch.

Download AudioSet data

Download AudioSet:

cd data/AudioSet
wget http://storage.googleapis.com/us_audioset/youtube_corpus/v1/csv/balanced_train_segments.csv
wget http://storage.googleapis.com/us_audioset/youtube_corpus/v1/csv/eval_segments.csv
wget http://storage.googleapis.com/us_audioset/youtube_corpus/v1/csv/class_labels_indices.csv

In these .csv files, there are URL links for each audio clip.

filenames.xlsx contains the names of the data that were additionally removed from the data in the training without music condition.

Quickstart

jupyter notebook

Select music kernel in the jupyter notebook.