Language specific LLMs, voice multi-modality