A paralyzed woman speaks again in the voice from her wedding video

The woman in the wheelchair cannot move her face. Beside her, propped where visitors can see it, an iPad speaks for her, in her own voice.

The researchers had a head start. They had video from her wedding day, showing how she sounded, how her voice rose and fell with feeling, and even a little about how she moved.

The work comes from Dr. Eddie Chang, a chair of neurosurgery and bioengineer who studies speech and language, and a childhood friend of Fei-Fei Li, the Stanford professor who leads Stanford's Human-Centered AI institute. Chang studies how patterns of activity in the brain become commands to the voice box and throat, and he has used that link to bring people essentially out of locked-in syndrome, a condition in which a person is conscious but cannot move or speak.

For this patient, who is paralyzed, that science meant she could speak through a computer. Li described how the team paired that decoded speech with an iPad placed next to her still face, and taught it with the wedding video so it carries her voice and her emotive patterns.

Li said the face on the screen matters more than the voice alone.

The result can hold a conversation, not just say words. The system is constantly updated through machine learning, a method in which software improves by finding patterns in new examples, and it pays attention to whom she is talking to and how they respond.

That responsiveness is the difference Li points to. A voice alone only says what she wants to say. A face on a screen that answers, adjusts, and interacts lets the world interact with her at a different level of depth.

Li called this embodiment. She acknowledged that it is still on a flat, two-dimensional screen, but she called it an exponential leap over accurate sound alone.

She remains in the wheelchair while the conversation happens on the screen beside her.