Dynamic Fusion Multimodal Network for SpeechWellness Detection