Fetch.Domovina.Tv

Create a comprehensive audio processing pipeline, orchestrated by a shell script, designed to take YouTube podcasts and turn them into structured articles. The process should start by refreshing podcast URLs, then downloading the audio using `yt-dlp`. Next, convert the audio to WAV format. I need to integrate several AI steps: first, generate a Whisper prompt using a local LLM API call, then transcribe the audio using Whisper, and finally, perform speaker diarization using `pyannote`. After getting the transcript and speaker labels, I want to optionally summarize it with Gemini, and finally, use Gemini again to generate a detailed, third-person article by first creating a thematic JSON outline and then writing the full article content based on that structure. The entire workflow should be manageable via a main shell script.