With the exception of psychologists, psychiatrists, neuroscientists, neurosurgeons, neural imaging experts, and a bunch of other brain-related professions, absolutely no one knows anything about the human brain. It is a weird tangle of inscrutable fleshy noodles filled with ganglia and synapses and god knows what else. It’s a confusing place. But it’s getting at least somewhat less confusing, with researchers from Columbia University making enormous steps in turning brain signals into audible, intelligible speech.
[jwplayer gvkTjsCy]
In a paper titled Towards reconstructing intelligible speech from the human auditory cortex published this week in the journal Scientific Reports, researchers led by Nima Mesgarani were able to take brain signals recorded in subjects listening to speech, pass it through a digital synthesiser, and have that speech recognised with a high rate of success.
While this isn’t the first time brain signals have been converted to speech, historically the speech produced has been of too low a quality to be of any use. According to the paper, their tests showed a 75% intelligibility rate for the synthesis of the numbers from zero to ten.
The results were achieved by taking the signal from the brain and using a deep neural network to generate the set of necessary parameters required by a vocoder to create matching audio output to the original audio that subjects heard. The subjects in the study were five patients that underwent additional “invasive electrocorticography” while already undergoing treatment for epilepsy.
If you’ve got a real head for science, you can read the entire paper right here. If, like me, you are aggressively confused and enraged by all this high-level science-talk, rest assured: it is extremely dope.
