eMusic - Project of neural networks system which recognise emotion of the given audio.

Bachelor's final project

Topic: "Application of machine learning algorithms in music emotion recognition"

Dissertation available at: https://misio.fis.agh.edu.pl/media/misiofiles/162daeb54191392a9cc3090f8ce28f58.pdf

General

Project uses Essentia library for audio content analysis available here:

http://essentia.upf.edu/

and

https://code.google.com/p/neurolab/

for neural networks.

Since Essentia Python bindings are written in Python2.7 whole project is also writtent using Python2.7

How to get?

git clone http://www.bitbucket.org/pawellll/inz.git

How to run?

If you're using PyCharm, just open it and run the project in IDE.

If not, open terminal, run main.py, but remember that it should be done from level of main folder of the project in order to have proper file paths to resources, otherwise it's not going to work.

Options

usage: main.py [-h] [-a] [-t] [-p] [-e] [-s FILE] [-n N]

Emusic

Arguments:

• -h, --help show this help message and exit
• -a Analyse songs during program execution. Otherwise it's going to load already processed songs if exist
• -t Train neural network during program execution. Otherwise it's going to load ready neural network if exists
• -p Plot comparison of neural network recognition and expected results. Use only with -e
• -e Evaluate neural network
• -s FILE Flag allows to analyse one song which is loaded and then analysed by neural network if any exist. Should be used without any other argument
• -n N Set amount of hidden neurons in net. By default 30

Resources

Dataset used: http://cvml.unige.ch/databases/emoMusic/

Resources folder contains:

• manual for data set which is used for developing and evaluating neural network
• documentation for Essentia library used in this project
• clips folder containing all audio files used in this project
• annotations folder containing information about songs and information about their valence and arousal parameters