Seol mar théacs é seo: Feature Fusion Based on Main-Auxiliary Network for Speech Emotion Recognition