Whisper

Whisper is the transcription model I use the most, by a lot. It runs on my own machine, so it doesn’t cost anything, and it’s in most of my pipelines by default.

It’s quick, and it’s not perfect. That’s fine for the stuff where I just need the general idea: putting chapters on videos, getting the rundown of a call, turning a voice memo into notes. Most of what I transcribe falls into that bucket.

When the words actually matter, like captions people are going to read on screen, I use Deepgram instead. It’s more accurate and I can configure it to get better results, but it costs money, so it only gets the jobs that need it.