The Voice from Her Wedding Speech

On her wedding day, Ann Johnson gave a speech of about 15 minutes. It was recorded.

Many years later, that recording would help give her voice back.

Ann taught math and physical education at a high school in Regina, Saskatchewan. She also coached volleyball and basketball. She had a baby with her new husband.

In 2005, at 30, she was playing volleyball with friends. A brainstem stroke struck.

She could no longer move any muscle. She could no longer speak. This is locked-in syndrome: the person is aware, but trapped in their body. At best, Ann could make sounds like "ooh" and "ah."

For seven years, she communicated with a letterboard. Her husband, she says, grew tired of translating. Then an eye-tracking system let her write about 14 words per minute. A conversation runs at 150 to 200.

Her daughter knew only one voice for her mother: the British-accented voice of her communication device.

In 2022, Ann joined a clinical trial run by the University of California, San Francisco (UCSF) and the University of California, Berkeley. She was its third participant.

Neurosurgeon Edward Chang placed a thin rectangle of 253 electrodes on the surface of her brain. They picked up the signals that once went to the muscles of her face and mouth.

An artificial intelligence learned to turn those signals into words. Ann tried to speak. On the screen, an avatar she had chosen spoke her sentences, with expressions such as sadness or surprise.

The researchers had recreated that voice from her wedding speech.

Eighteen years after her stroke, Ann heard her own voice again.

In a feedback survey, she wrote that hearing it made her emotional.

In August 2023, the team published its results in the journal Nature. The system reached nearly 80 words per minute, with a 1,024-word vocabulary. It got it right about three times out of four.

It is not yet an everyday tool. The implant is connected by a cable to computers. In February 2024, it was removed, for a reason unrelated to the trial.

The research goes on. In March 2025, in the journal Nature Neuroscience, the team presented a new method. The delay between the attempt to speak and the sound dropped from about eight seconds to about one. Ann, who stays in touch with the researchers, said she felt in control.

A machine does not replace a voice. But it can give someone back their place in the conversation.

Her dream is to become a counsellor in a physical rehabilitation facility, and to talk with clients through a neuroprosthesis. She says why:

We become someone through others, and speech is one of the paths that connect us. Here, technology replaced nothing: it reopened that path.