Baum-Welch training example

Question

I'm the author of a new Baum Welch trainer using MapReduce for the Apache Mahout project (https://issues.apache.org/jira/browse/MAHOUT-627)

I'm looking for an example with a reasonably small data set which I can use to demonstrate the usage of the trainer I have developed.

Does anyone here know of any online demonstrations for this training which go beyond the normal "toy" examples? I'm interested in typical HMM training application areas--bioinformatics, NLP, Signal Processing etc.

Answer 1

A simple example from bioinformatics that is often used to teach students how to train a HMM with Baum-Welch are the CpG islands:

http://en.wikipedia.org/wiki/CpG_island

The problem is pretty straightforward when you have sequences of CpG islands and some reference sequences, and should be solvable in short time.

Answer 2

This isn't my field, so take this with a grain of salt.

I do know, however, that Hidden Markov Models play a huge role in speech recognition applications. Here's a very highly-cited paper, which I believe is the basis for a lot of current methods. I also know that the most common benchmark data set for this is called TIMIT, although I think there are better ones now.

Hope this is useful.