RAVDESS

The RAVDESS dataset, which stands for Ryerson Audio-Visual Database of Emotional Speech and Song, is a widely used collection of audio and video recordings of actors portraying various emotional states. The dataset was created to facilitate research in speech emotion recognition and related fields. It contains 7356 recordings, each lasting between 4 and 6 seconds. The recordings are spoken words and sung vowels performed by 24 actors, 12 male and 12 female.

Each actor provided recordings in 10 different emotional states: calm, happy, sad, angry, fearful, disgusted, surprised, neutral, happy-song, and sad-song. For each emotional state, three different expressions were used, and two different re-enactments were performed. The dataset includes audio recordings, video recordings of the actors' faces, and corresponding transcriptions.

RAVDESS has been instrumental in the development and evaluation of machine learning models for identifying emotions in speech. Its rich and varied content allows researchers to explore different acoustic and visual features associated with emotional expression. The dataset is freely available for research purposes, making it accessible to a broad community of scientists and developers. Its standardized structure and comprehensive annotations contribute to reproducible research and facilitate comparisons between different emotion recognition systems.