For about $16 a week, I think running a real sample of Bray audio through Universal 3.5

Pro plus diarization plus identification alongside your current Whisper Firenode results would

be an exceptionally worthwhile experience.

Yeah, that's a bit more than I want to spend when I consider that I'm comparing it with

running my own GPU server, which costs me nothing.

I already own it.

Just electricity, which is not cheap, but it's nowhere near that much.

So let's look at this another way.

If I use Assembly AI but didn't use the recognition option, how much is it cheaper per minute?

And then also, tell me about this NVIDIA option.

I'm open to switching away from Firenode if I'm not using Whisper, or even I could use

Whisper and this NVIDIA option, but I need to know how much...