[Feature Request] Support transcription of audio files larger than 25mb

[aschmelyun]

OpenAI's Whisper API has a hard limit of 25mb per upload. As it is right now, Subvert doesn't split up uploaded files, and just sends the entire audio file to Whisper.

Will need to implement a way of batching file uploads to prevent exceeding this limit.

This will require refactoring a large part of the video processing section and kind of includes #18 as well.

For now, try to limit uploads to around 22 minutes unfortunately.

[kuubus]

Same Problem.
To solve the problem, it would be necessary either to transcode to meet the requirements or to split into segments. In case of segments, the most likely error could be caused by the transition from one segment to the next. Another possibility would be to split the audio file into segments with a slight time offset and clean up the transition in the delivered text.

[ma-lalonde]

A simple workaround in the meantime, if I may suggest, would be to check the audio file size and automatically compress it as needed. I pre-process the audio with the command:
ffmpeg -i video.webm -vn -acodec mp3 -fs 26M output_file.mp3