Learning Physiology-Informed Vocal Spectrotemporal Representations for Speech Emotion Recognition

Researchers developed PhysioSER, a new AI model that dramatically improves speech emotion recognition by analyzing the physiological features of human voices. Unlike standard models that only use amplitude, it couples both amplitude and phase data informed by vocal anatomy. The compact, plug-and-play system was tested across 14 datasets in 10 languages and 6 AI backbones, showing superior interpretability and efficiency. It has already been deployed on a humanoid robotic platform for real-time interaction.