This research introduces a novel approach for multilingual speech emotion recognition without requiring training data in target languages.

• Combines contrastive learning with large language models to refine speech features across languages
• Enables zero-shot recognition of emotions in previously unseen languages
• Addresses challenges of voice characteristic variations and linguistic diversity
• Creates more robust speech emotion recognition systems for multilingual environments

For linguists, this breakthrough enables development of emotion-aware language technologies that work across cultural and language boundaries without needing extensive language-specific training data.

Large Language Models Meet Contrastive Learning: Zero-Shot Emotion Recognition Across Languages