Pošalji tekstualnu poruku: Feature Fusion Based on Main-Auxiliary Network for Speech Emotion Recognition