Comparative Reasoning: Making an Audio Language Model Better at Comparing Emotions
For researchers in speech emotion recognition and audio-language modeling, this work provides a data-efficient approach to ordinal comparative reasoning, though it is incremental as it combines existing techniques (LALMs, GeMAPS, DPO) in a new way.
The paper introduces a reasoning-guided ordinal speech emotion recognition framework that enables large audio-language models to perform comparative emotional judgments between two utterances. The method achieves improved preference prediction while requiring only 5% of the training data used by conventional systems.
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.