Comment by millions

Has anybody seen a comparison of actual speech-to-text capabilities, preferably done in a proper manner (using speech recordings of different accents, etc)?