Meta said it has made significant progress in itseffort to create more realistic AI-generated speech systems.

The systems are held back because they can only learn from written text.

“),” the team wrote in the blog post.

Someone chatting with a smart speaker while working on a laptop in a home office setting.

Luis Alvarez / Getty Images

But Utnik said that most current speech recognition platforms are only 75-80% accurate.

Correction 04/05/2022: Corrected the spelling of Ryan Monsurate’s name in paragraph 9.

A young child interacting with a smart speaker at the family dinner table.

Oscar Wong / Getty Images