Experimental programs that use speech recognition to perform
real-time language translations have been kicking around for years now, but
Microsoft took the idea to a new level at its TechFest 2012event last week. Like other translators, the software developed by Microsoft
Research allows you to talk to it in your native tongue and send it out the
speaker of a device as another language -- Spanish, French, Chinese, or such. What
comes out of that speaker, though, isn't the ersatz speech of a computer robot,
but an ersatz approximation of what you sound like. What's more, the software
will create a 3D image of your head that makes it look as if you're speaking
the translation. Called Monolingual TTS, the system currently has 26 languages
in its repertoire. Acquainting the system with a voice does take more time than
it does in your typical speech recognition program -- about an hour of training
Monolingual TTS in your vocal tones. If the application can be stuffed into a
smartphone, it would be a boon to international travelers. Language training is
also a good fit for the technology. Microsoft's vision in the translation space
is broader than what has been produced by its rivals. It wants to use lifelike
virtual avatars that not only mimic your looks but also your voice and the
movements of your lips when you speak. The results produced from that vision
can be impressive, but they can be creepy,
too. Google, which has a widely-used online
text-based translator, has also
been delving into speech-based translation.

No comments:
Post a Comment