Breaking Resource Barriers in Speech Emotion Recognition via Data Distillation