Three research teams recently made progress in turning data from electrodes surgically placed on the brain into computer-generated speech. Using computational models known as neural networks, they reconstructed words and sentences that were, in some cases, intelligible to human listeners.