We use the Viterbi training algorithm (Durbin et al., 1998) to find an optimal sequence of system states and optimize the transition matrix A as well as the parameter matrix.
Once the CNF model is trained, we can calculate the optimal alignment using the Viterbi algorithm (Viterbi, 1967).
Before training the compact model, a state level segmentation is made using the Viterbi algorithm with the backend model, i.e., the segmentation specifies the alignment between the states and the frames of the training data.
These probabilities are decoded using the Viterbi algorithm.
And then, the total bits are decoded by using the Viterbi soft decoding algorithm.
The stretch where they match is aligned with the audio using the Viterbi algorithm.
It can be determined in polynomial time using the Viterbi algorithm [21].
Each model produces a probability value by backtracking through model states using the Viterbi algorithm.
Next, we predicted constrained elements by using the Viterbi algorithm.
For PRED-TMBB, the predictions were performed using the Viterbi and Posterior Decoding algorithms.
