Improving Speech Emotion Recognition Through Cross Modal Attention Alignment and Balanced Stacking Model

TL;DR

This paper introduces a multimodal speech emotion recognition system that uses cross-modal attention and balanced stacking to improve performance on naturalistic speech data, addressing class imbalance and fusion challenges.

Contribution

It proposes a novel architecture combining cross-modal attention and ensemble stacking with balanced training strategies for improved emotion recognition.

Findings

Achieved MacroF1 score of 0.4094

Attained accuracy of 0.4128 on 8-class SER

Demonstrated effectiveness of balanced stacking ensemble

Abstract

Emotion plays a fundamental role in human interaction, and therefore systems capable of identifying emotions in speech are crucial in the context of human-computer interaction. Speech emotion recognition (SER) is a challenging problem, particularly in natural speech and when the available data is imbalanced across emotions. This paper presents our proposed system in the context of the 2025 Speech Emotion Recognition in Naturalistic Conditions Challenge. Our proposed architecture leverages cross-modality, utilizing cross-modal attention to fuse representations from different modalities. To address class imbalance, we employed two training designs: (i) weighted crossentropy loss (WCE); and (ii) WCE with an additional neutralexpressive soft margin loss and balancing. We trained a total of 12 multimodal models, which were ensembled using a balanced stacking model. Our proposed system achieves a MacroF1 score of 0.4094 and an accuracy of 0.4128 on 8-class speech emotion recognition.