CREMA-D

7,442 acted audio-visual emotional clips from 91 demographically diverse actors speaking 12 sentences in 6 emotions at 4 intensity levels.

• Task: emotion, speaker
• Languages: en
• Domain: acted emotional
• License: ODbL 1.0 (database) + DbCL 1.0 (contents)
• Homepage: https://github.com/CheyneyComputerScience/CREMA-D
• Paper: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4313618/

Recommendation

Good for demographically diverse acted emotion recognition and audio-visual affect work, with crowd-sourced perceptual ratings included. Choose it when speaker/ethnic diversity matters; emotion is acted and the corpus is modest in size.

Getting the data

Obtain from the dataset homepage.

Openly available on GitHub under Open Data Commons licenses.

Suggested processing

A recommended VoxKitchen pipeline ships in the repository at examples/pipelines/emotion-recognize.yaml — run it with vkit docker run.