In my experience, medium is often the sweet spot for English accuracy vs speed, especially if following-up with a post-processing pass. The large options are all fine, but can severely slow it down. There are some speed checks on my website if you're curious (link not posted because I don't want to hijack another post's app).

I also got hammered when it came to diarization... I found that the biggest pain was creating an appropriate environment for cross-compatibility of the different backends required for whisper/faster-whisper/pyannote. It's especially challenging on older systems, so major kudos for giving it a shot.

Have you gotten any traction yet from the community?

1. I literally just put together a personal "Hunter" dashboard to help identify posts and topics that are relevant to my interests. The hunter is basically just a RSS crawler for keywords, scoring for relevant topics, and similar. It brought me here! Given that it was quick to toss together, I'm not surprised that other folks DM me on Reddit with "Hey, I found you..." spam.

2. A specialized app that consolidates unstructured audio into longer-term project notes like a project manager (check out my website if interested, it's free). I wanted something private/local that helped support people who’d rather talk through their work and end up with usable project history — status updates, decisions, blockers, progress over time — without hand-crafting every note.

3. Last week I polished my website and made a new "Observatory" landing page (https://www.watchlightstudio.com/)! It's a cool little space theme that shows folks what I'm working on, has dust/particle effects, and similar. I'm still finding my brand, and this was a fun opportunity to dip my does into .css

Faster-whisper does allow for VAD tuning via API. I don't think it's exposed natively for Whisper though.

What's your diarization pipeline? Pyannote?

I'd taken a different approach that used a LLM clean-up pass to summarize and progressively compress the transcript for ultra-long content, but I like the idea of targeted "pay attention here" flags.