The Baum-Welch algorithm is an algorithm for
estimating the parameters of the HMM from training data.
The HMM is a complete statistical model for
the series of measurements
and therefore
defines the probability density function
. Therefore, once the
parameters have been determined, it is easy to use the HMM
as a classifier. Furthermore, it is also easy to generate
"typical" measurement sequences. This aspect of the HMM has
always fascinated me since in principle, it would be possible
to train an HMM on a specific human speaker, then
create totally random “jibberish" that sounded like the
same speaker. I have always wondered if certain politicians
are already using such a device. For further information on HMM's,
the reader is referred to the tutorial by Rabiner [65].