A fake word conveniently lacking google results.
Setup.sh is intended to be run on a clean VM in my home XCP-ng server. You probably don't want to run it on your own machine as-is. If this works I'll make it cleaner and more compatible.
The goal is to use WhisperX diarized transcriptions of a single spoken audio corpus (in this case, youtube channels inserted into shared/workspace/urls_list.txt) along with some type of agentic pipeline for entity extraction, to reliably figure out the ground truth transcript names. If that fails I plan to get weird with multimodality. Llava-1.6-34b may be involved.
