The combination of Large Language Models (LLMs) with automatic speech recognition (ASR) is transforming NLP tasks. This research examines their integration using extensive open-source Chinese datasets:

• Data and Model Exploration: Detailed testing of various configurations of speech encoders and LLMs to determine their impact.
• Advancements in Training: Introduction of a three-stage training approach to better align auditory and textual information.
• Achieving State-of-the-Art Results: Achieved top performance on the AISHELL1, TestNet, and TestMeeting test sets.
• Public Accessibility: Release of scripts and models for the community, fostering reproducibility.

Incorporating LLMs into ASR not only pushes the boundaries of technology but also provides valuable insights into optimizing performance using specialized datasets. The public release of resources promotes further experimentation and advances in the field.