Unveiling the Potential of LLM-Based ASR on Chinese Open-Source Datasets