Open The-Gupta opened 6 years ago
Hi Gupta, the RAVDESS is ideal for emotion recognition projects. It's a validated, multimodal database of emotional speech & song, released under a Creative Commons license. The 7,356 recordings were produced by 24 professional actors in a neutral North American accent. Speech contains 7 universal emotions (calm, happy, sad, angry, fearful, surprise, disgust), and song contains 5 emotions. Each emotion is produced at two levels of intensity, with an additional neutral expression. Files are available in audio-video, video-only (no sound), and audio-only formats. High levels of perceptual validity were reported from 319 raters.
Download the RAVDESS from Zenodo.
PLoS One paper describing construction & validation.
This repo. provides only 47 short audio files with valence and arousal annotation CSV files. Could someone suggest a larger and Open Source dataset preferably in English* for training the regression model?!
*I know emotions are universal, even then I feel it would be better to train on English dataset for the desired application