杏吧原创

Nine years ago, a brain-stem stroke left Erik Ramsey almost totally paralysed, but with his mental faculties otherwise intact. Today he is learning to talk again 鈥� although so far he can only manage basic vowel sounds.

In 2004, Ramsey had an electrode implanted in his speech-motor cortex by Philip Kennedy鈥檚 team at , a company based in Duluth, Georgia, US, who hoped the signal from Ramsey鈥檚 cortex could be used to restore his speech.

Interpreting these signals proved tricky, however. Fortunately, another team headed by at Boston University, Massachusetts, US, has been working on the same problem from the opposite direction.

Guenther and his colleagues have used information from brain scans of healthy patients to monitor neural activity during speech. These studies show that the brain signals don鈥檛 code for words, but instead control the position of the lips, tongue, jaw and larynx to produce basic sounds.

Guenther鈥檚 research group then developed software that could recognise and translate the patterns of brain activity during speech.

Quick learner

When they teamed up with Kennedy, they could use their software to interpret the signals from Ramsey鈥檚 implanted electrode and work out the shape of the vocal tract that Ramsey is attempting to form. This information can then be fed to a vocal synthesiser that produces the corresponding sound.

The software is now translating Ramsey鈥檚 thoughts into sounds in real time, so Ramsey hears his 鈥渧oice鈥� as he makes a sound, effectively bypassing the damaged region of his brain stem.

This gives Ramsey immediate feedback on his pronunciation, which he can use to rapidly hone his speaking skills in the same way infants do when learning to talk. Initially, when prompted to produce a vowel sound such as 鈥渆e鈥� or 鈥渙h鈥�, he hit the correct sound around 45% of the time. Over the course of a few weeks, Ramsey鈥檚 accuracy has risen to 80%.

Tricky consonants

In the future, Guenther says that the goal is to give Ramsey the ability to speak complete words with fluency, but that will require software and hardware improvements.

鈥淭he synthesiser is good for vowels, but not for consonants,鈥� he says. 鈥淲e鈥檙e going to move to a synthesiser with better consonant capabilities, where the patient would have more control over jaw height, for instance.鈥�

But a more complicated system would be harder to control, says Guenther. 鈥淩ight now, Erik is controlling two dimensions to create vowels,鈥� he says. 鈥淏ut for consonants he would need seven dimensions 鈥� three to control tongue movement, two for lip movement, and one each for jaw and larynx height.鈥�

at the Technical University of Berlin in Germany is intrigued by the study, but wonders whether it will come at a price for Ramsey.

Migraine risk

鈥淚f you鈥檙e trying to help someone to communicate with the external world, the question is how strenuous and stressful is the interface,鈥� he says. 鈥淪ome brain control interfaces can lead to migraines.鈥�

Ramsey still retains control of one muscle in his eye, through which he can communicate. Although that communication route is very slow, M眉ller thinks it might produce more data given sufficient time.

He also thinks that Guenther and Kennedy鈥檚 technique might not be suitable for fully 鈥溾�� patients, who can鈥檛 control any muscles.

鈥淚f you have a completely locked-in patient then maybe the brain begins to degrade because they鈥檙e not able to control anything,鈥� he says. 鈥淭hose patients might be completely unable to communicate even using a brain-control interface.鈥�

Guenther presented his latest results at the in France on 3 July.