Get the latest tech news
A unified acoustic-to-speech-to-language embedding space
This study links acoustic, speech and linguistic data with brain activity during real-life conversations to create a model that predicts neural responses during speech with high accuracy.
Four patients (2 females, gender assigned on the basis of medical record; 24–53 years old) with treatment-resistant epilepsy undergoing intracranial monitoring with subdural grid and strip electrodes for clinical purposes participated in the study. The ECoG preprocessing pipeline mitigated artefacts due to movement, faulty electrodes, line noise, abnormal physiological signals (for example, epileptic discharges), eye blinks and cardiac activity 47. To account for intersubject variability, we analysed time points of the neural encoding peaks with linear mixed models (LMMs), including a random intercept per patient using restricted maximum likelihood estimation.
Or read this on Hacker News
