MERaLiON-SER: Robust Speech Emotion Recognition Model for English and SEA Languages

TL;DR

MERaLiON-SER is a robust speech emotion recognition model that effectively captures both categorical and dimensional emotions across English and Southeast Asian languages, outperforming existing models and enhancing empathetic audio systems.

Contribution

The paper introduces MERaLiON-SER, a novel multilingual speech emotion recognition model using hybrid loss functions for joint discrete and dimensional emotion modeling, with superior cross-lingual performance.

Findings

Outperforms open-source speech encoders and Audio-LLMs in multilingual settings.

Effectively captures both emotion categories and intensity, valence, dominance.

Demonstrates robustness across English and Southeast Asian languages.

Abstract

We present MERaLiON-SER, a robust speech emotion recognition model designed for English and Southeast Asian languages. The model is trained using a hybrid objective combining weighted categorical cross-entropy and Concordance Correlation Coefficient (CCC) losses for joint discrete and dimensional emotion modelling. This dual approach enables the model to capture both the distinct categories of emotion (like happy or angry) and the fine-grained, such as arousal (intensity), valence (positivity/negativity), and dominance (sense of control), leading to a more comprehensive and robust representation of human affect. Extensive evaluations across multilingual Singaporean languages (English, Chinese, Malay, and Tamil ) and other public benchmarks show that MERaLiON-SER consistently surpasses both open-source speech encoders and large Audio-LLMs. These results underscore the importance of specialised speech-only models for accurate paralinguistic understanding and cross-lingual generalisation. Furthermore, the proposed framework provides a foundation for integrating emotion-aware perception into future agentic audio systems, enabling more empathetic and contextually adaptive multimodal reasoning.