This project is based on the papers contained in this folder

Summary of the dataset:

• each recording is 40 min long
• which is composed of 2 recordings of 20 min
• 10 songs of 2 min make one recording

The data is of the h5 format

We manually find with script.py the beginning of the music in the data Start of music for Louis = 34532 Start of music for Charles = 21189

Then dataset_generator.py cuts the 40 min recordings into the 20 songs. Each 20 min recording is an array of 10 songs. Each song is a dictionary of the signals. The keys are the zones.

You can access the recordings later by calling

from dataset_generator import *

Our choice was to use multitapering for the FFT of the signals. For more details ask Charles Masson

See the scripts unsupervised.py and unsupervised_rythmic.py The frequency powers of the 10 songs are plotted in order to look patterns in the data. The resulting plots are in the folder images.

Sklearn is used to try to see if we can predict if a song is rythmic or not. The script is also interactive in the command prompt.