The goal of the project is to perform speech emotion recognition on audio files. The dataset in use contains both songs and speech, by 24 distinct actors performing 8 different emotions.
All the code is organized inside src where you can find:
src/loader.py: loading of raw data;src/data_preprocessing.py: feature extraction from raw data;src/dataset.py: dataset management, such as splits, scaling, etc;src/models.py: collection of functions that define or train models;src/utility.py: plots and initial study on the dataset.
The study is divided among different notebooks:
01_exploration.ipynb: initial study on the dataset;02_training.ipynb: initial training of neural networks and simpler models;03_data_augmentation.ipynb: experiments with data augmentation.04_cross_validation.ipynb: cross validation results.