An Embarrassingly Simple Approach for LLM with Strong ASR Capacity