If you haven't, you should see the speech to text of gboard on a pixel phone, it is accurate, fast, available in every language.

On a computer download handy or a similar app.

There are so many extremely fast extremely accurate FOSS AI speech to text models only 50-200 megabytes. Made by NVidia, Google, Microsoft, Chinese firms, all open source and free.

We have FUTO on android, which is using a custom fork of ancient slow and heavy Whisper from OpenAI which nobody uses anymore.

I wish someone with enough knowledge would build a modern speech to text based on these new models. And win a noble prize for it.