Отправить по sms: Feature Fusion Based on Main-Auxiliary Network for Speech Emotion Recognition