This project is part of a master's thesis for the: course DA231X - Degree Project in Computer Science and Engineering, Second Cycle 30hp
These scripts are meant to load and transform datasets into Dataframes for ease of use (that is, no actual data, just preparation scripts). The data is loaded by reading file_names from the corresponding dataset-directories (see below how for directory structure).
Install with pip:
pip install git+https://github.com/felixlut/SER_dataloader.git@main
The package is meant to be accessed via python like this:
from dataloader import Dataloader
dataset_name = 'cremad' # /cremad/emodb/...
data = Dataloader(data_args.data_path, force_reload=False) # After the first instance of this command, the dataset is cached. force_reload=True for circumventing the caching
dataset_as_df = data.load_dataset(dataset_name)
The data is supposed to be stored in .../datasets/{dataset_name}
, and should have the below described structure overall.
.
├── datasets
├── dataset_name
├── wav/wav # 2 layers of wav, kaggle datahandling is weird...
| ├── file_abc123.wav
| └── ...
├── wav_telephone/wav_telephone # same as for wav, 2 layers cuz kaggle...
| ├── file_abc123.wav
| └── ...
└── meta_data.csv # Possibly extra csv:s containing some metadata used for constructing the dataframes, ex. actor_2_gender.csv
The following datasets are supported, and have to have the same directory name mentioned below (don't mind the inconsistency of the data
postfix, kaggle is once again coming through with only the highest quality of data-handling). These can also be derived from the correspponding scripts.