Training the Model

We learn the HMM probabilities from a training set of [signal,words] pairs. Current systems are 80-98% accurate, depending on the hardware and the size of the vocabulary.