Text this: A Lightweight Tri-Stream Feature Fusion Network for Speech Emotion Recognition