So we had a choice: either we let scientists at the University of Sheffield attach electrodes to our tongues, or they were going to go and build their own artificial mouth. Because we're not so into the whole electrode thing, they built "Anton," an animatronic tongue made of soft silicone to help them understand speech and subsequently improve speech-recognition software. This isn't the first of its kind, believe it or not -- there's much competition in the robotic mouth world. Because speech recognition systems aren't really benefiting from simply crowding them full of recorded speech, researchers want to better understand how the mouth produces sound and then create algorithms that can simply recognize speech patterns rather than try to match recordings to recordings. Sounds about right to us. Peep the creepy video after the break.

We know, it's a dream of yours to one day hear what Neanderthals from eons past sounded like when they conversed of DAPs possessing a few kilobytes of storage space and longed for computers that wouldn't take up entire caves. Okay, so maybe that's a stretch, but an anthropologist at Florida Atlantic University has seemingly figured out a way to actually recreate what ancient human speech (probably) sounded like. By utilizing reconstructions of a trio of Neanderthal vocal tracts, the crew was able to engineer what a spoken "E" would've sounded like via a computer synthesiser; in time, they're hoping to construct an entire Neanderthal sentence. Of course, not everyone's convinced that the discoveries made thus far are entirely accurate, but be sure to hit the read link a take a listen for yourself (hint: it's nothing like those Bud Light commercials).

[Via Slashdot, image courtesy of ABC]

Truth be told, we've already seen instances where technology has enabled individuals to speak without speaking, but a brain implant placed into Eric Ramsey's head could certainly raise the bar in this field. The wireless electrode, which resides just six-millimeters below the surface of his brain, records pulses from 41 surrounding neurons involved in speech generation, and thanks to software developed by the associated team, those thoughts will hopefully be translated into words in the not-too-distant future. Currently, the group feels that they can accurately identify the word Ramsey is thinking up 80-percent of the time, and in the coming weeks, a computer will begin giving the patient real-time feedback so he can perfect the art of thinking out loud. 'Course, the team responsible isn't likely to be satisfied until an unadulterated conversation can take place, but it seems we're well on our way to seeing that come to fruition.

[Via BBC]

Apparently, not only will our robotic overlords be capable of crushing and / or dismembering us with their razor-like talons and arm-mounted machine guns, but they will also be able to perfectly mimic our oral movements when speaking. According to reports, researchers from the Tokyo University of Science and the Musashino Red Cross Hospital have created a robot capable of reproducing tongue and lower jaw movements associated with speech. The prototype -- a model of the mouth -- is able to perfectly copy the movements needed to produce vowel sounds. Using a silicon tongue, which is supported by three aluminum scaffolds and wired to eight pneumatic muscles, and a lower jaw which pivots on an axle, researchers are able to program the "mouth" with data based on magnetic resonance imaging. The next step, they say, is to create an artificial mouth which covers all movements (including consonants), allowing the bot to be used for speech therapy, foreign language training... or the basis for Terminator heads. [Warning: read link requires subscription]

Quite a lot has transpired in the android universe in the past two years, and it's fairly safe to say that Kansei has made a few solid strides during that time as well. A Japanese research team has purportedly crafted a working prototype that can "pull up to 36 different facial expressions based on a program which creates word associations from a self-updating online database of 500,000 keywords." According to a professor at Meiji University's School of Science and Technology, the idea of the project is to "create a flow of consciousness in robots so that they can make the relevant facial expressions," and the device relies on 19 movable parts and a silicon face mask in order to work its magic. Developers also noted that speech abilities should be added within a few years, and while we doubt you had to guess, it's quite likely that fully developed Kanseis will one day roam nursing homes as Japan seeks to care for its quickly growing geriatric set.

While we English speaking folk may prefer our pocket translators to start in English and convert to some other foreign tongue, Casio's latest rendition helps those fluent in Japanese order tickets to the Space Needle, haggle at the Chevrolet dealership, and of course, pick up the correct ingredients for warm apple pie. Aimed at middle / high school students, the XD-SW4800 lineup of handheld dictionaries reportedly house the denotation and correct pronunciation for over "85,000 languages," which ironically enough, probably translates to "85,000 words" when not read through a broken translator. Nevertheless, these handy gizmos come in a variety of colors, and aside from supporting keyboard input, it can recognize stylus-written characters and toss back definitions while pronouncing the word to you via headphones. If America isn't your final destination, it also comes with five other major languages on CD that can be loaded on the 50MB of internal space or on your SD card if necessary, and while these things won't last forever, the approximate 60 to 130 hours of battery life should be more than enough to get you acclimated. So while you're waiting for your passport to show up, make sure you pick up Casio's latest travel necessity -- if you can manage the admittedly steep ¥47,250 ($394) to ¥52,500 ($437) price points, that is.

[Via Impress]

We can call it the real-time audio obscenity mask, or we can call it the autobleep, the cursekiller, whatever. The point is Microsoft's patented a unique system for bleeping out (actually just munging or silencing) combinations of phonemes it identifies as obscenities in audio streams without making use of time shifting, or otherwise requiring manual human intervention to make sure the FCC doesn't stick  them with some harsh fines. Would this theoretical system be able tell the difference between the real f-bomb and funk, or even homonym swears like those synonyms for a donkey and female dog? You know the ones. Tell you what, let's put it to work on a satellite radio hip-hop station or an old George Carlin record and see how we do.

[Via Ars Technica]