Hidden Markov Models (HMMs) are a popular approach for speech recognition. Commonly, a left-to-right Markov chain topology is used, where each phoneme is represented by a sequence of states (typically three) [