Labelling is a data-preprocessing technique that adds meaningful label(s) to raw data so that machine learning models can learn from.
In supervised learning, labelling is a pre-requisite to produce training data and each piece of label have to done manually
AWS Recognition
In unsupervised learning models, the labels will be genered by an algorithm and might not be user readable. Ground Truth - A piece of training data serves as objective truth for the model to make inference objectively, the better your ground truth the better your labelling