Hacker News Reader

This is exactly part of this Google USM approach, although these pretrained models are significantly smaller than ChatGPT. They reference this paper [1] which contains more details on the pretrained text-only alignment with the speech model.

[1] https://arxiv.org/abs/2209.15329

Use whisper. It doesn't get confused in that way.

Are you sure about that? I've seen YouTube captions understand homophones, and Google Assistant definitely can (though that may be linked to some other system).

Surely the speech recognition model itself learns some basic language statistics in order to recognise homophones?

There are no "pure speech" models. All ASR uses language models.

Same with spell check, with context and phonetics you could make autocorrection better than any human.