Creating feature data for training.

To test the tools, we need to generate HMM output data from the above-defined model. Execute the script file software/hmm_example.m. The program calls the function software/hmm_maketestdata.mwhich generates the 2-dimensional feature data as described above. The call is

    [x,istart,nsamp]=hmm_maketestdata(Pi,A,nrecord,nsteps,N,NFEAT);

There are 10 records of length 400 segments, thus x is size 2-by-4000. The auxiliary outputs istart,nsamp are vectors containing the starting samples and lengths of each of the ten records. This makes it possible to locate individual records within the matrix. The script then plots the data using the command

	plot(x(1,:),x(2,:),'b.');
	xlabel('MEAN');
	ylabel('STDV');

and waits for keyboard input. The resulting figure is shown in Figure 13.21.

Figure 13.21: Scatter plot of the HMM output features. The three states can be seen individually. Compare the plot with the table of means and standard deviations.