Last updated on April 16, 2014 at 8:44 EDT

Reconstructed Speech

February 1, 2012

These are frequency spectrograms of the actual spoken words (top) and the sounds as reconstructed by two separate models based solely on recorded temporal lobe activity in a volunteer subject. The words — Waldo, structure, doubt and property — are more or less recognizable, even though the model had never encountered these specific words before.  Credit: Brian Pasley, UC Berkeley