Ever since the first Terminator movie in 1984, Terminator cyborgs have had the ability to duplicate the voice of any given human they hear, an ability deployed again in last night episode of Terminator: The Sarah Connor Chronicles, when our plucky band of heroes has its cell phones intercepted. It’s not so far fetched — pretty much this exact scenario has been worrying real security researchers for some time.
In 1997, Australian researchers developed a system that could fool automated voice verification systems that authenticate users based on their voiceprint. By 2001, our future-overlord-enablers at AT&T had created software called Natural Voices that could copy any human voice.
The current non-killing-machine value of this type of software is for realistic text-to-speech applications—for instance the ability to listen to emails using a cell phone, without that note from your Aunt Jemima sounding like it comes from a Dalek, or for use with the screen readers that visually impaired people use with their computers. Combined with speech recognition systems, speech synthesis already allows some automatic language translation applications — in the future, the addition of the ability to mimic a given voice could allow you to conduct, say, a conservation in real time over the telephone in which your speech is translated into another language in your own voice.
Going beyond sneaky terminators, this would be something like a basic version of the universal translators many sci-fi shows use to get over the awkward fact that all the aliens seem to speak perfect, if somewhat North-American accented, english…
September 24th, 2008 at 11:40 am
I really doubt we could ever achieve real-time. It’s not a technical but a linguistic problem. But on the fly translation with a lag of a few seconds is not too far away, I think.
September 24th, 2008 at 7:30 pm
[...] Terminator: At Least Cyborgs Enunciate – Discover’s Science Not Fiction blog looks at the real-life science behind “The Mousetrap”. [...]
November 12th, 2008 at 6:02 pm
Krystian has a point, for example:
Dogs are smart. (Human)
Smart are dogs. (Yoda)
Say that the syntax (ie, the spelling & definitions of the individual words) is the same, and the only job of the interpreter is to correct the semantics (ie, the order & grammar of the words). How could this be done in real time? In this scenario it would seem one has to wait till all the words of the short sentence have been spoken in order to reverse all the words of the sentence into the other being’s language. And this is just a short sentence. In some languages, perhaps a whole paragraph has semantics which differ from another language’s. (Or a whole movie…. ever watch “Memento”?)