"Gaussian Mixture Model for Mandel / Ellis algorithm" This class holds
the features needed for the Mandel Ellis algorithm: One full covariance
matrix, and the mean of all MFCCS.
This method performs the maximization step for this component given the
sample points and the estimates p_ij = P(C=i | x_j) for sample
j of being generated by this component under the assumption that this
sample has been drawn from this GMM.