Correction of ASS file content

[gormson]

Hello,

This is at the very edge of a problem, that's why I open an issue, so sorry if this is not appropriate :)

When using the /transcribe-media endpoint to generate the .ass file, there are sometimes wrong transcriptions, especially for the names or technical data. 99% is working perfectly but sometimes these wrong identifications are popping out.

I tried a turnaround via some chatGPT steps to correct it, but even if working with SRT files, the format of .ass files makes it really random on the results.

At the end I can only use .srt files that I can correct easily but without taking the benefit of the fancy captions format with .ass files.

Considering for the use cases that sometimes the voice over is scripted, do you think this would be possible to provide this script with the /transcribe-media payload? So that the extraction could directly be self corrected?

Thank you.

[stephengpope]

We are changing the way the modules work so that you can produce an SRT file with transcribe-media, modify it, and then use that to generate the .ass file and fancy captions directly, instead of having to modify the .ass file.

Is that what you mean?