RAVDESS

Acted emotional speech and song from 24 professional actors across 8 emotions at two intensity levels (1,440 speech audio files).

• Task: emotion, speaker
• Languages: en
• Domain: acted emotional
• License: CC BY-NC-SA 4.0
• Homepage: https://zenodo.org/records/1188976
• Paper: https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0196391

Recommendation

A clean, balanced, widely-used benchmark for speech emotion recognition — good for reproducible, controlled experiments and quick baselines. Emotions are acted (not naturalistic), the corpus is small, and the non-commercial license prohibits commercial use.

Getting the data

Obtain from the dataset homepage.

Open access on Zenodo; non-commercial license.

Suggested processing

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