WESR-Bench is an expert-annotated natural speech dataset with word-level non-verbal vocal events, featuring both discrete events (standalone, denoted as [tag]) and continuous events (mixed with speech, denoted as ...).
Discrete events (15):
inhale, cough, laughs, laughing, crowd_laughter, chuckle, shout, sobbing, cry, giggle,exhale, sigh, clear_throat, roar, scream, breathing
Continuous events (6):
crying, laughing, panting… See the full description on the dataset page:
https://huggingface.co/datasets/yfish/WESR-Bench.
