Hidden Markov ModelHMM
Definition
A statistical model that represents sequences of observations as transitions between hidden states.
Hidden Markov Models were the dominant approach for speech recognition for decades before deep learning. In an HMM-based ASR system, each phoneme or word is modeled as a sequence of hidden states, and the acoustic signal at each time step is treated as an observation emitted by the current hidden state.
The Viterbi algorithm finds the most likely sequence of states (and thus the most likely transcript) given the observations. While HMMs have been largely replaced by neural network approaches in state-of-the-art systems, the mathematical framework influenced many concepts still used in modern ASR, including CTC alignment.