High-order HMM in Matlab Files

This package implements duration high-order hidden Markov models (DHO-HMMs). We also show that the DHO-HMM can be reduced to the hidden semi-Markov model (HSMM) and hidden Markov model (HMM) by tying some parameters of the DHO-HMM.
 
Before you start to use the programs, you should first prepare the training and testing data. Excerpts of TIDIGITS database can be obtained from http://cronos.rutgers.edu/~lrr/speech%20recognition%20course/databases/isolated_digits_ti_train_endpt.zip and http://cronos.rutgers.edu/~lrr/speech%20recognition%20course/databases/isolated_digits_ti_test_endpt.zip. The root directory for the training data, isolated_digits_ti_train_endpt, and the root directory for test data, isolated_digits_ti_test_endpt, should be placed under the "wav" directory so that we can run the Matlab scripts without modification. You can start with “main_all.m” or “main_all_mp.m”, which will in turn invoke the feature extraction program, generate the list of training and testing files, and then invoke the training and recognition programs. 

To prepare your own data, you can modify the Matlab script file "dr_wav2mfcc_e_d_a.m" for extracting the feature vector sequence from your own waveform data. You also need to create a .mat file containing a list of training data and another .mat file containing a list of test data, where the first field of a record in the list represents the word ID (in integer) and the second field is the path of the data file. Example Matlab script files for creating training and testing list files are "generate_selected_TI_isolated_digits_training_list_mat.m" and "generate_selected_TI_isolated_digits_testing_list_mat.m", respectively. 
In script files “main_all.m” and “main_all_mp.m”, you should modify the system parameters for your use. The meaning of these parameters are as follows.

HOSTATE_NO_RANGE: The number of high-order states in a model. It can be an positive integer. It can also be a range of positive integers so that you can obtain the recognition rate for different state number.
HMM_ORDER_RANGE: The order of dependency. It can be an positive integer. It can also be a range of positive integers so that you can obtain the recognition rate for different order of dependency.
MODEL_NO: The number of models.
dim: the dimension of the feature vector.
TR_ITERATION_BEGIN: the begin of training iteration, which is normally set to 0.
TR_ITERATION_END: the end  of training iteration.
TS_ITERATION_BEGIN: the begin of test iteration.
TS_ITERATION_END: the end  of test iteration.

The following parameters are for advanced users who would like to trace and modify the programs. 
INITIALIZATION_METHOD: You can also choose one of the 4 initialization methods for the initialization of the DHO-HMM.
min_frame_count and min_self_transition_count are used to obtain smoothed estimation of the transition probabilities.