Impact of annotation modality on label quality and model performance in the automatic assessment of laughter in-the-wild.