Emission probabilities are a component of Hidden Markov Models (HMMs) that represent the likelihood of observing a particular output from a specific hidden state. They are crucial for decoding sequences and are typically estimated using training data or algorithms like the Baum-Welch algorithm.