Hidden Markov models in practice
Hidden Markov models for sequence labeling

Whatever the data or signals you wish to exploit with hidden Markov models, you first need to format them. This pre-processing and feature extraction can be quite complex, and can benefit from strong a priori knowledge of the signals, as is the case in automatic speech recognition.

For example, figure 12 shows the process by which a writing signal is pre-processed into the input of a Markov system. The image of the word (or phrase) is sliced into small windows by dragging a narrow window from left to right. For each window position, a number of features are calculated. For example, we can divide the...