quentinverlhac/pianocktail

Using Deep Learning for emotion analysis in songs

Music emotion recognition

Using Deep Learning for emotion analysis in songs.

Inspired by the work of Jan Jakubik, Halina Kwasnicka, Music Emotion Analysis Using Semantic Embedding Recurrent Neural Networks.

Dataset

Download datasets and add them in the data folder, matching paths defined in config.py.

Emotify dataset can be found here. Download it and store it under data/emotifymusic.

Trainning models

• Choose the model in config.py
• Run python processing.py to process the data

To train with validation:

• Run python split.py --validate to generate train/validate datasets
• Run python train.py {model_name} --validate to train the selected model on the train dataset and validate it

To train and test:

• Run python split.py to generate train and test datasets
• Run python train.py {model_name} to train the selected model on the train dataset without validating it
• Run python test.py {path_to_model} to test the model saved at path_to_model on the test dataset

Requirements

Make sure you have ffmpeg installed locally.

• sudo apt install ffmpeg on Ubuntu.
• brew install ffmpeg on macOS.

pip install -r requirements.txt

To update requirements.txt: pip freeze > requirements.txt