Kmeans: Classify a feature sequence as a m_mixnum cluster, calculating the center (means) and variance (Var) of each m_mixnum cluster, and the proportion of each cluster (prior).
GMM: The Kmeans computed m_mixnum cluster, the center and variance of the Gaussian function is initialized with the center and variance computed by Kmeans, and the center, variance and weight of each cluster are adjusted according to the Gaussian function.
Iteration Cutoff criteria: The average of the log (p) of the sequence is less than 0.001.
Use a m_mixnum Gaussian function to represent a state.
CHMM: The characteristic sequence of a gesture is divided into statenum states, each state is divided into several eigenvalues, and all eigenvalues that belong to this state are trained with a mixed Gaussian model, that is, Statenum≠m_mixnum. That is, each state is represented by several Gaussian functions.
Gmm-chmm Simple Grooming