Better text export [feature request]

Been using transcribee to evaluate and edit podcasts for clients, and would love a better format for exporting.
Maestra.ai has a nice schema for Word I tried to replicate in a python script (linked, at bottom) that I would love to see in Transcribee.

Sadly I don't have the knowledge to add to your project, as I managed to make this script with chatgpt. It takes the VTT-file and process it into the desired output.

Basically, the format output of this code merges the speakers into paragraphs and gives the time code for the entire paragraph.

Example:

00:00:00.680 --> 00:00:12.800
Speaker X:
Lorem ipsum dolor sit amet, consectetur adipiscing elit. Nunc rutrum sapien sit amet tortor varius luctus. Vestibulum id lorem faucibus, commodo tellus ac, imperdiet massa.

00:00:12.939 --> 00:00:25.940
Speaker Y:
Fusce tristique libero mauris, a malesuada lectus feugiat eu. Sed metus tellus, tincidunt a magna vel, hendrerit varius enim. Integer in lectus finibus, fermentum tellus et, iaculis dolor.

https://gist.github.com/phixofor/f5286ca2c42a2339cfe43bec322efe2a

Start by reviewing the linked gist and the current VTT export behavior. Compare the requested Word-oriented schema with the example, focusing on merging consecutive speaker segments into paragraphs and assigning each paragraph an overall time range. Done means the export produces the requested speaker-labelled paragraphs and timestamps.

Written by the indexing model from the issue text.