A compact numerical representation of data (like audio) created by a neural network, capturing its essential features in a much smaller form than the original.