Thursday, October 02, 2008

Thought to Speech

Amazing. [Link]

In the largest room of the dark, cluttered office, tables are stacked with computer monitors and electronics equipment, and a web of cables drapes between dislodged ceiling tiles. In the center of the room, Erik Ramsey is sitting in his wheelchair, wearing a blue sweat suit and slippers, with a bundle of wires coming out the back of his head. He's staring at a wall onto which Kennedy has projected a matrix of six words: heat, hid, hat, hut, hoot, and hot. They represent each of the major English vowel sounds. Kennedy, tall and stately at sixty, asks Erik to think about making the sound uh-ee. As he does, a green cursor jitters across the wall from hut to heat, and a booming vibrato pours out of the speaker: "uuuhahuuuuhaheeeeeeee." The sound is coming straight from Erik's brain.

Kennedy is trying to help Erik become the first human being ever to have his thoughts translated directly into speech. In November 2004, Kennedy's team put Erik into an fMRI scanner and showed him images of animals. While the scanner monitored Erik's brain activity, Kennedy asked him to say the name of each animal in his head: "This is a lion. This is an elephant." The fMRI produced a map guiding them to the precise area of Erik's brain that was activated when he tried to speak, a region of the premotor cortex that controls movement of the mouth, lips, tongue, and jaw. A few weeks later, neurosurgeons working with Kennedy opened Erik's skull and threaded a tiny glass cone containing three long, hair-thin Teflon-coated gold wires into exactly that part of his brain.

Within a few months, Kennedy had Erik producing short words like dada and mama, but the model he was using to interpret the brain signals was too simple ever to decipher real speech. When Kennedy further analyzed the signals, he discovered that he could detect thirty-two of thirty-nine English phonemes, or basic sound units, in the electrode's output. In 2006, Frank Guenther, a computational neuroscientist at Boston University, joined the project. An expert in the brain's speech systems, Guenther helped Kennedy develop a computer decoder that could turn those patterns into a prosthetic voice.

No comments:

Post a Comment