

Advanced enough to recognise any words it has been trained on.
For example, these offline models can recognise commonly used, conversational words, with a high degree of accuracy(99% in general usage). Because they have been trained on those words. It will get more inaccuracies when trying to recognise unfamiliar scientific/technical words.
Whisper+, FOSS offline voice-recognition.

For regular daily speech, used by someone without a strong accent, yes, more than sufficient.