Music data analysis

Project description

The goal of this project is to determine the genre of a given music based on data related to it using data mining and machine learning tools. From the GTZAN dataset, we use a CNN that trains with the image representation of music (spectrogram).

How to use

Requirements

• Python 3.7.x or Python 3.8.x
• TensorFlow
• Scikit-Learn
• Librosa
• Pydub
• Numpy
• Matplotlib
• Seaborn
• Pickle
• Pandas
• Optional: NVIDIA Cuda (for training on the GPU)

Build

As the project already contains a trained CNN, the Training part is not necessary. You can go directly to the General use (predictions) part if you wish.

Training

First, you need to download the GTZAN dataset at this address:
http://marsyas.info/downloads/datasets.html (click on Download the GTZAN genre collection)

Once the archive has been downloaded, run:

tar -xvf [path_to_gtzan_archive]genres.tar.gz -C[path_to_project_root_directory]
cd [path_to_project_root_directory]
rm -rf *.mf

The third command allows you to delete useless files from the archive (that are not music). The project tree should look like this:

etc/
genres/
	blues/
	classical/
	[...]
	rock/
src/
.gitignore
README.md

For a CNN to be effective, it needs a large amount of data. To do this, you will split the music into several 4 second chunks with an overlap of 1 second. For this, just run this script in the root directory:
python src/split_music.py

You now need to generate the dataset on which your CNN will train.
Still in the root directory, run :
python src/data_generation.py (Be patient, this may take a while)
N.B. : The script will generate three files: two pickle files containing the spectrograms and labels for the CNN and a CSV file containing calculated variables (MFCC, RMSE, ...). The latter is not needed for the CNN, so you can skip this step by commenting out line 112 in src/data_generation.py: save_calculated_variables_in_csv(music_directory)

Now you will train and save the CNN and display the accuracy curve and the confusion matrix. For this, run in the root directory:
python src/CNN.py
(Be patient, this may take a while)

General use (predictions)

You can use the CNN in a concrete case. To do this, just run in the root directory:
python src/test_predictions.py [path_to_music_to_classify]
N.B. 1: The music must be in .wav format. If this is not the case, you can use Convertio.co or any other tool of your choice to convert your music to the correct format.
N.B. 2: Make sure that the file name does not contain any spaces. E.g.: my_music.wav

You will see the genre predicted by the CNN in the terminal.

Team members

• Prof. Christophe ROSENBERGER, teacher and tutor
• Nathan MICHEL, project manager
• Raphaël ANCETTE, AI developer
• Camille GUIGNOL, AI developer
• Arnaud RIO, mobile developer