Emotion Datasets

A large number of emotional speech datasets have processing scripts and metadata available in the ERTK repository. Currently the scripts only deal with audio data, but in future we intend to support video data.

Currently supported datasets:

• AESDD

• ASED

• BAVED

• CaFE

• CMU-MOSEI

• CREMA-D

• DEMoS

• EESC

• EMO-DB

• EmoFilm

• EmoryNLP

• EmoV-DB

• EMOVO

• ESD

• eNTERFACE

• IEMOCAP

• JL-corpus

• MELD

• MESD

• MESS

• MLendSND

• MSP-IMPROV

• MSP-Podcast

• Oréau

• Portuguese

• RAVDESS

• SAVEE

• SEMAINE

• ShEMO

• SmartKom

• SUBESCO

• TESS

• URDU

• VENEC (Public subset). The full dataset is available on request.

• VIVAE

Standardising datasets

Each dataset has a subdirectory in the datasets directory. This subdirectory contains a process.py script, which takes a path to the original dataset, and converts it to a simple standard format, consisting of annotation CSVs and audio files. The script also generates a corpus.yaml file, which contains metadata about the dataset.

You can run the process.py script for a dataset, such as EMO-DB, as follows:

cd datasets/EMO-DB
python process.py /path/to/EMO-DB