Friday, April 6, 2012


Microsoft wants to turn monolingual speakers into multilingual speakers by computerizing their voices and accurately recreating them in other languages, intonation and accent intact.
Wesley Fenlon
Microsoft Research wants to help you speak Chinese, and all you have to do is spend about an hour speaking into your smartphone--in English. The group's latest prototype, a universal translator, needs about an hour of voice time to build a sample. After an hour of voice analysis, your intonations and pronunciation have been computerized; they're then matched against a reference speaker in another langauge and merged into a new voice. The result is meant to synthesize your voice into Chinese or Spanish or any of 26 supported languages.
Microsoft's ultimate goal is perfect human voice parity--a smartphone-produced voice that sounds exactly like your own, but speaking a language you don't understand. For now, the resulting voice is noticeably computer-generated, but Microsoft Research may be pulling ahead of similar translation services like Google Translate.

No comments:

Post a Comment