Comparative Reasoning: Making an Audio Language Model Better at Comparing Emotions

2026-06-22

AI Summary

Existing large audio language models exhibit limited performance in pairwise voice comparison tasks across dimensions such as emotion, environment, language, prosody, and interpersonal context. This work proposes a reasoning-guided ordinal speech emotion recognition framework that conditions the audio language model on paired speech inputs and integrates semantic audio descriptions with GeMAPS acoustic features to generate interpretable reasoning trajectories. By incorporating Direct Preference Optimization (DPO), the approach enhances the model’s ability to discriminate subtle emotional differences. Notably, this is the first study to jointly leverage reasoning trajectories and DPO for emotion comparison in audio language models. The method achieves substantially higher accuracy in emotion preference prediction while requiring only 5% of the training data used by conventional approaches, thereby reducing data dependency and improving model interpretability.

Abstract

Large audio-language models (LALMs) can reason about audio, yet it remains unclear whether they can perform comparative judgments between two speech signals along emotional, environmental, linguistic, prosodic, and interpersonal dimensions. We study this question in the context of speech emotion recognition (SER), where the model determines which utterance exhibits higher arousal, valence, or dominance. We introduce a reasoning-guided ordinal SER framework that conditions an LALM on paired speech inputs. The model is trained using reasoning traces generated from both semantic audio descriptions and acoustic evidence derived from GeMAPS features, enabling interpretable comparative decisions. Beyond direct supervision, we also employ direct preference optimization to encourage stronger separation for emotional differences. Experiments show that the proposed framework improves preference prediction while requiring only 5% of the training data used by conventional ordinal SER systems.

Problem

• comparative reasoning
• audio language models
• speech emotion recognition
• ordinal judgment
• emotional comparison

Innovation

• audio-language model
• comparative reasoning
• ordinal speech emotion recognition
• reasoning trace
• direct preference optimization