If you weren’t entirely certain whether I said “One year ago” or if I actually said “When you’re a goat” – even if it sounded more like the latter – you would of course assume the former because it makes more sense, right? Right. Then, you’re a human, not voice recognition software.
What I really don’t get is why some kind of logic cannot be built into voice recognition software that would do a better job at recognizing when the interpretation is likely incorrect based on its nonsensical nature. Why couldn’t the software algorithm compare its best guess vs. one of its alternate guesses, and pick the most likely? Then, this kind of error would be so much less common:
What I actually said to my computer: Serving Tulsa
How my voice software translated it: Serving tolls suck
I mean, come on, which of these phrases is likely the most common or reasonable, right? Another example:
What I actually said to my computer: structural process
How my voice software translated it: trucks rule process
Maybe trucks do rule. But a structural process surely is a commonly used phrased. Maybe in the future, Dragon NaturallySpeaking will have this kind of delineation honed. Meanwhile, watch what you say to it.