Iteratively re-aligning the model with the data and updating the model parameters based on the single best alignment.
Token passing
This is a really nice way to understand, and to implement, dynamic programming in HMMs.
The Viterbi criterion
At the core of the Viterbi algorithm is this simple dynamic programming operation: discard all paths that cannot win.
The Viterbi algorithm is dynamic programming
When we apply dynamic programming to an HMM, we call it the Viterbi algorithm.
Relating DTW to the HMM
The template in DTW is now replaced by a model, but otherwise the methods are conceptually very similar.
Dynamic time warping
This is just a taster of what we’ll cover in the main lecture, but do watch it so that you start to understand some of the key concepts.
This is the new version. Still under construction.