Microsoft shows off universal translator that preserves your voice

Microsoft has been showing off a lot of these lab projects lately. I guess they think it helps with their marketing. The problem is they've not just started working on these types of projects. Their R&D has always been doing stuff like this, but either they never seem to come out in products, or if they do, it happens a decade later. So other than marketing bragging rights, this is all pretty pointless if they don't show up in products at all, or not for a decade.

I believe something similar was developed at CMU under the name ABBY. I wonder if they are the same; nevertheless, the CMU tech was impressive - real-time translation with voice characteristic preservation. We wanted to use it to develop a plugin for Skype for international business, but it didn't pan out. A missed opportunity in my opinion. However, now that Microsoft owns Skype, maybe we can expect an announcement soon.

The samples in the article don't do a very good job of demoing the technology.<p>You can download an mp3 of the talk: <a href="http://msrvideo.vo.msecnd.net/rmcvideos/160725/dl/160725.mp3" rel="nofollow">http://msrvideo.vo.msecnd.net/rmcvideos/160725/dl/160725.mp3</a> forward to 19:25 for a demo of english and mandarin<p>the full explanation of the system starts at 12:00

<i>For example, Microsoft’s standard model of Spanish will have a default “S” (ess) sound, but the training process replaces it with your “S” sound. This is done for every individual sound (phoneme)</i><p>I don't think it can work as a phoneme for phoneme replacement; there must be a different heuristic at work.<p>Different languages have different phonemes and one of the most telling things of a non-expert non-native speaker is that they don't get the sounds right.<p>Eg "E" is pronounced differently in English, German and Czech. Japanese speakers don't differentiate between "L" and "R". Spanish "C" is in some contexts different to "C"s in other languages. Etc.<p>Obviously you can't pronounce a sound that you don't even know exists.<p>I would love to hear more examples.

Impressive technology. But this doesn't really solve the problem of accurately translating from one language to another, with all the ambiguities of language, does it? I mean, as far as I can tell, the really hard part is completely transferring the meaning of something spoken/written in one language into another. Sometimes it's obviously just not possible (like certain jokes, etc.). Most of the time though, it's just very hard to do with software.

Makes me think of Iron Man, when Gweneth Paltrow clicks on the 'Translate' button in the terrorist video, and it was the original voice but translated into English. I never thought it would actually be possible, but this looks pretty interesting.

This technology, once perfected, will usher in a new era of fraud and crank calls.

I don't understand the samples in the article. The English audio says something, and the Spanish audio translates to "Welcome to TechFest 2012, were today you will see first-hand how Microsoft Research is studying the key technological tendencies that will define the XXI century." which is completely different.

Listening to the two audio samples in the report, I would say the voices don't sound very similar and the translated one sounds very mechanical.

While the technology is very impressive, as a native Chinese/English speaker, I can tell you that the mixture of Chinese and English demonstrated is only useful for bilingual speaker like myself.<p>Also watching the demo of this presentation in contrast with Apple's keynote I can really see a world of difference on how two companies communicate with their customers.

(Pet peeve) This is not a universal translator! Until you show me a translator that can learn a <i>new</i> language, you're just building a common multi-language translator.

I know it's Microsoft, but why are they using Windows Media for web video in 2012? I don't want to install Flip4Mac just to watch the demo.