GENERAL keeps the acting intention.
The original scene and delivery instructions preserve what a character shows, hides or lets slip. They describe the requested performance, not verified internal emotion.
SCRIPT follows the recorded audio.
Recognized words, sentence durations, pauses and described burst intervals come from ASR and the audio annotation models. The separate observed caption lets you compare intention with prediction.
Search the acting directions
BM25 searches GENERAL, including the original delivery progression. Terms use OR matching; combinations receive a relevance score. For an emotion or quality ranking, choose “Selected predicted score”. A z-score is a standardized value, not a probability.
Scope: completed top-ups and extreme acting cases. The older core collections remain available in the dataset.