Used primarily for speech recognition, a TDNN uses delay elements to scan across time steps and compute larger and more abstract features. 27.07.2023 17:54 aior