It's not that straightforward. I would say that the most important part of pre-processing is how to break the transcript into parts.
But there are many more things to improve. It's a pipeline: you can add more models, you can train them, you can change prompts, you can post-process the results. But it takes time. More time each step.