Music Genre Detection Project

Automatic music genre classifiers

Usage

Installing prerequisites

The project is built in python2.7

Conda install

Create and activate a fresh virtual environment:

conda create -n [yourenvname] python=2.7 anaconda
source activate [yourenvname] 

Install required packages:

conda install -n [yourenvname] --yes --file requirements.txt

If you get a PackageNotFoundError, add necessary channels:

conda config --add channels conda-forge
conda config --add channels litl-rnd

and run the previous command again.

Pip install

In a fresh virtual environment, run

pip install -r requirements.txt

Dataset

For training and testing our classifiers we use GTZAN music genre collection, but we are restricted to a subset which is not including classical and jazz genres, because one of our classifiers depends on lyrics, and classical and jazz music representatives are mostly instrumentals. To download raw data, run

python download_data.py

This will create data/ directory in project's root, and download and extract both audio and lyrics datasets into it's subdirectories: data/genres/ and data/lyrics/

After that, run

python preprocess_data.py

to create preprocessed datasets (also stored in data/ directory).

Training

To train CRNN (Convolutional Recurrent Neural Network) classifier, run

python train_crnn_model.py

To train Naive Bayes classifier, run

python train_naive_bayes_model.py

Models and weights are saved in models/ directory.

App server

This project includes a small python app that demonstrates the CRNN classifier.

If you want to enable downloading audio from YouTube, make sure you have youtube-dl and pafy libraries installed:

pip install youtube_dl==2017.5.23
pip install pafy==0.5.3.1

Command to run demo app:

python server.py

Now you can access the demo app from your browser at localhost:8080.

There are some example songs at /examples

Scripts

• melspectrogram.py - generates mel-scaled log-power spectrogram
• lbr_melspectrogram.py - generates mel-scaled log-power spectrogram using librosa lib

Usage:

python melspectrogam.py path/to/audio/file
python lbr_melspectrogam.py path/to/audio/file

• get_lyrics.py - azlyrics, metrolyrics and oldielyrics crawler

Usage:

python get_lyrics.py 'The Smiths' 'Handsome Devil'  

• nb_predict.py - uses Naive Bayes model to predict song genre

Usage:

python nb_predict.py path/to/lyrics/file 

or

python nb_predict.py 'The Smiths' 'Handsome Devil'  

Presentation

• English: https://docs.google.com/presentation/d/1HUuCpDaassL3Gv5E0909chKlsM4lW4hI-D0pcd43uy8/edit?usp=sharing
• Serbian: https://docs.google.com/presentation/d/1T7h60enALTi2MvlT5r0MIeifv8tQiTAlrOHrlZa7wxk/edit?usp=sharing