WATCH: How AI gave a paralyzed woman her voice back

Decoding the signals of speech

Ann was a high school math teacher in Canada before her stroke in 2005. In 2020, she described her life since in a paper she wrote, painstakingly typing letter-by-letter, for a psychology class.

“Locked-in syndrome, or LIS, is just like it sounds,” she wrote. “You’re fully cognizant, you have full sensation, all five senses work, but you are locked inside a body where no muscles work. I learned to breathe on my own again, I now have full neck movement, my laugh returned, I can cry and read and over the years my smile has returned, and I am able to wink and say a few words.”

As she recovered, she realized she could use her own experiences to help others, and she now aspires to become a counselor in a physical rehabilitation facility.

She learned about Chang’s study in 2021 after reading about a paralyzed man named Pancho, who helped the team translate his brain signals into text as he attempted to speak. He had also experienced a brainstem stroke many years earlier, and it wasn’t clear if his brain could still signal the movements for speech. It’s not enough just to think about something; a person has to actually attempt to speak for the system to pick it up. Pancho became the first person living with paralysis to demonstrate that it was possible to decode speech-brain signals into full words.

With Ann, Chang’s team attempted something even more ambitious: decoding her brain signals into the richness of speech, along with the movements that animate a person’s face during conversation.

To do this, the team implanted a paper-thin rectangle of 253 electrodes onto the surface of her brain over areas they previously discovered were critical for speech. The electrodes intercepted the brain signals that, if not for the stroke, would have gone to muscles in Ann’s lips, tongue, jaw and larynx, as well as her face. A cable, plugged into a port fixed to Ann’s head, connected the electrodes to a bank of computers.

For weeks, Ann worked with the team to train the system’s artificial intelligence algorithms to recognize her unique brain signals for speech. This involved repeating different phrases from a 1,024-word conversational vocabulary over and over again until the computer recognized the brain activity patterns associated with all the basic sounds of speech.

“It was exciting to see her go from, ‘We’re going to just try doing this,’ and then seeing it happen quicker than probably anyone thought,” said Ann’s husband, Bill, who travelled with her from Canada to be with her during the study. “It seems like they’re pushing each other to see how far they can go with this.”

Rather than train the AI to recognize whole words, the researchers created a system that decodes words from smaller components called phonemes. These are the sub-units of speech that form spoken words in the same way that letters form written words. “Hello,” for example, contains four phonemes: “HH,” “AH,” “L” and “OW.”

Using this approach, the computer only needed to learn 39 phonemes to decipher any word in English. This both enhanced the system’s accuracy and made it three times faster.

“The accuracy, speed and vocabulary are crucial,” said Sean Metzger, who developed the text decoder with Alex Silva, both graduate students in the joint Bioengineering Program at UC Berkeley and UCSF. “It’s what gives Ann the potential, in time, to communicate almost as fast as we do, and to have much more naturalistic and normal conversations.”

Adding a face and a voice

To synthesize Ann’s speech, the team devised an algorithm for synthesizing speech, which they personalized to sound like her voice before the injury by using a recording of Ann speaking at her wedding.

“My brain feels funny when it hears my synthesized voice,” she wrote in answer to a question. “It’s like hearing an old friend.”

She looks forward to the day when her daughter — who only knows the impersonal, British-accented voice of her current communication device — can hear it too.

“My daughter was 1 when I had my injury, it’s like she doesn’t know Ann … She has no idea what Ann sounds like.”

The team animated Ann’s avatar with the help of software that simulates and animates muscle movements of the face, developed by Speech Graphics, a company that makes AI-driven facial animation. The researchers created customized machine-learning processes that allowed the company’s software to mesh with signals being sent from Ann’s brain as she was trying to speak and convert them into the movements on her avatar’s face, making the jaw open and close, the lips protrude and purse and the tongue go up and down, as well as the facial movements for happiness, sadness and surprise.

“We’re making up for the connections between her brain and vocal tract that have been severed by the stroke,” said Kaylo Littlejohn, a graduate student working with Chang and Gopala Anumanchipalli, Ph.D., a professor of electrical engineering and computer sciences at UC Berkeley. “When Ann first used this system to speak and move the avatar’s face in tandem, I knew that this was going to be something that would have a real impact.”

An important next step for the team is to create a wireless version that would not require Ann to be physically connected to the BCI.

“Giving people like Ann the ability to freely control their own computers and phones with this technology would have profound effects on their independence and social interactions,” said co-first author David Moses, Ph.D., an adjunct professor in neurological surgery.

For Ann, helping to develop the technology has been life changing.

“When I was at the rehab hospital, the speech therapist didn’t know what to do with me,” she wrote in answer to a question. “Being a part of this study has given me a sense of purpose, I feel like I am contributing to society. It feels like I have a job again. It’s amazing I have lived this long; this study has allowed me to really live while I’m still alive!”