Preprocessed IEMOCAP Data for the Paper "Taming Ambiguity: A Unified Approach for Aleatoric and Epistemic Uncertainty in Speech-Text Emotion Recognition"

This dataset contains the preprocessed IEMOCAP features for both the speech and text modalities. Each modality dataset includes the train, validation, and test sets. The properties for each modality dataset are as follows: Speech: WavLM-large, 6 s, Original + Gaussian Noise + Speed Perturbation + Reverberation (Stacked) Text: RoBERTa-Base, 64 tokens, Original + Back Translation + Contextual MLM (Stacked) The IEMOCAP dataset itself is available upon request from https://sail.usc.edu/iemocap/.