The technology that enables you to transcribe your work conference may help the paralyzed person speak again.
Researchers at UC Berkeley and UC San Francisco use generated AI to reduce the delay in people with severe paralysis when they try to speak with computer devices playing sounds. Their work helped a woman named Ann, who suffered a brainstem stroke at the age of 30 in 2005 to get close to real-time communication. Ann’s voice sounded like her own, as the model trained her recordings from before the stroke.
UC Berkeley Ph.DDD Cheol Jun Cho said the deployment of AI has been deployed in several different ways, allowing researchers to improve neural prosthesis, which could take longer. Students in electrical engineering and computer science and co-leading authors of the study, which emerged in March Natural Neuroscience.
Here is an example of generating AI tools – using the same basic techniques Chatbot Joe told me that like Openai’s Chatgpt and Anthropic’s Claude or Google Meet are helping medical and scientific researchers solve problems that may take longer to solve. AI experts and supporters point out the use of the technology in medicine A huge areais it there Design new drugs Or provide Better testing and diagnosis.
“AI is accelerating the pace,” Joe said. “Sometimes we imagine the timeline will be a year or two. Now, the pace is about three years.”
Cho said the technology that helped ANN is a proof of concept, but it shows a path toward a tool that might be more plugged in the future.
Speed up the speech
The problem with existing neural prostheses is delay. There is a time when the person begins to try to speak with the true origin of sentences and hear when. Joe said the previous technique meant An had to wait until one sentence was completed before starting the next sentence.
Ann saw in his first study in 2023, able to communicate through computers that read signals her brain is trying to send to the muscles that control speech.
“The main breakthrough here is that she doesn’t have to wait until the sentence is finished,” he said. “Now, whenever she is going to speak, we can actually stream the decoding program.”
The prosthesis includes a series of electrodes implanted on the surface of the brain and connected to the computer library via cables. It decodes the control signal Ann’s brain sends to the muscles that control speech. After Ann chose the words she intended to say, an AI reads these signals from the car’s cortex and gives them life.
To train the model, the team tried to say a sentence on the on-screen prompt. They then used data from that activity to map signals in the motor cortex and used Gen AI to fill the gap.
Cui said the team hopes the breakthrough will lead to a scalable and easier access device.
“We are still working to make it more accurate and have a lower latency,” he said. “We are trying to build something that can be plugged.”
Using AI to transform from thought to speech
Joe said the team used AI in several different ways. One is to copy An’s hurt voice. They used pre-injury recordings to train a model that could produce her voice.
“She was so excited when she heard her voice again,” Joe said.
https://www.youtube.com/watch?v=mgsokggbbxk
The biggest change is real-time transcription. Cho compared it to a tool when a transcription of a presentation or meeting took place.
This work is built on 2023 Research This uses AI tools to help ANN communicate. This work still has a great delay in when Ann tries to speak and when words are produced. The study was greatly delayed, and Ann told the team that it felt more natural.
“She reported that she embodies her speech,” Joe said.
