3 ms·
What an awesome idea. I wonder how well this technique would work for different data types. Text and audio for example.
by zardo 9y ago
What an awesome idea. I wonder how well this technique would work for different data types. Text and audio for example.
- DonaldPShimoda 9y agoMost audio translation is first transcribed to a text representation, I think, and then that text is what's translated. If you assume a perfect speech recognizer (for the sake of argument), then the end translation should be the same. Or am I misunderstanding your point?
- zardo 9y ago>Or am I misunderstanding your point? Only that you are talking about specifics, while I was talking in generality. (IMO) unsupervised learning of semantic relationships is a 'big deal'. This seems to be a very general method that would extend to any pair of data streams, but that hasn't been proven out yet.