2 min

Researchers from Meta developed an AI model that decodes speech from thought. The technology allows people to produce words without speaking.

When humans indulge in conversations, they think about how words will form sentences in their head. Thinking of words is enough for your brain to create activity. You don’t necessarily need to speak. Different parts of the brain are responsible for how the mouth works. Some areas play a significant role in speaking and understanding languages.

Researchers from Meta’s Facebook AI Research Labs have invented an AI algorithm that may change the lives of millions of people, especially those who suffered neurological traumas that affects their ability to communicate, type or gesture.

“We’ve developed an AI model that can decode speech from noninvasive recordings of brain activity”, said FAIR Labs research scientist Jean Remi King, “Decoding speech from brain activity has been a longstanding goal of neuroscientists and clinicians, but most of the progress has relied on invasive brain-recording techniques.”

AI clears noise and decodes speech

Many of us are acquainted with brain scans like CT and MRI, which produce comprehensive brain activity images. But they only focus on structures and ignore activities. The ways to get a hold of brain actions, such as skull apertures and electrode placements, have all been invasive until now.

Non-invasive technologies such as EEG and MEG have the power to scan the brain from outside and record activities without any operation. However, most fail to provide a clear picture of brain activities due to noise.

To counteract this challenge, scientists employed machine learning and AI algorithms to clean those noises through a model, Wave2vec 2.0, an open-source tool designed in 2020 by FAIR scientists.

“Given a snippet of brain activity, it can determine from a large pool of new audio clips which one the person actually heard”, King added. “From there, the algorithm infers the words the person has most likely heard.”

It’s clear that AI can decode noisy recordings of perceived speech. Now, Meta is working on translating brain activities without audio recordings. As a result, advanced technologies and new ways of controlling machinery just by thinking words can be developed.

Tip: Meta AI resolves background noise issues with AV-HuBERT