Ask HN: Most accurate ML speech-to-text API?

[lumens]

I'm building a project that relies on at least pretty-good transcription with timestamps for each word and ideally speaker diarization.

Right now I'm using Google Cloud's Speech-to-Text, but the accuracy is underwhelming when transcribing a Zoom call (50%ish).

Am I likely to fare much better with Azure/AWS? What about Symbl.ai?

[taf2]

Which model are you using on the zoom calls? Also are you used enhanced or just default? There a lot of factors with any engine.

[mdrabla]

While sometimes more expensive, I've found GCP the best option (from an accuracy standpoint) for STT diarization