I saw whisper recommended, but I was curious how it compares to the other robust ASR systems (like nvidia's Nemo + Riva). I found this Twitter thread that seemed relevant: https://nitter.net/lunixbochs/status/1574848899897884672
Long story short, it depends on what you want to use it for. Different models and different training sets can help optimize for different things. Also, if you're in a domain with very uncommon speech patterns (think doctor shorthand or radio lingo), you'll need to understand how difficult it will be to customize generated models to do better in your space. I think Nemo + Riva does well at this; but I'm not as familiar with other options.