S2Dialog: Multimodal Dialogue Retrieval with Semantic and Acoustic-Style Modeling

S2Dialog implements a unified framework for dialogue-level retrieval by decoupling semantic and acoustic modeling into dedicated textual and acoustic retrievers. The architecture moves beyond utterance-level matching to capture global semantic coherence and stylistic consistency across entire multimodal dialogue sequences.