[celestialcheese]

whisper-1 has this with the verbose_json output. Has word level and sentence level, works fairly well.

Looks like the new models don't have this feature yet.