Yeah, I heard about that, I doubt it will beat universal translator I am working on.
Funny thing, I was planning on including klingon too
So far my experience with speech recognition and automatic translators has been underwhelming.
I am talking about google products.
Neural nets are great but they are still very dumb in the sense that if you allow complete freedom they can produce pretty unexpected results. And this is what I observed in both translator and speech recognition. These things are sometimes remarkably adequate but then they can be remarkably inadequate too, in other words very spotty.
And standard NN metric of not understanding what is going on looks impressive but in reality you never understand it with NN, even with primitive ones, and it is not a good thing because you are not getting necessary insight to move forward. I mean you say a word and it recognizes it hilariously wrong, then what? Do you blindly retrain the nework? Retraining will fix this problem but breaks other things. And this is what I see in speech recognition - hilariously unexpected errors.
Anyway, my speech recognition promise to be much more robust and versatile.
