HERest.

HERest.

HList.

HRest.

HMM model, file.

Link to the developer of "htkconvert".

http://www.liacs.nl/~erwin/

Problem solved!

I made a mistake in the hmm model file (the transition matrix, and exactly in the start state). The probability start_state - start_state must be 0, and start_state-first_real_stat must be 1. check the attached file.

I wanna know about deep detail of this,

Can you show me your format of dataset that you use,please

Thank you :)

Hello,

As I mentionned, the problem that I asked here is already solved.

About the dataset format, I extract some features from the image of each character and I transform these features to a discret sequence (a sequence of integers). So, like this, each character of the alphabet will have a set of discret sequences. I used these sets to initialize each discret HMM (I modelize each character by a discret HMM). After this initialization, it comes what it's called "embedded training", where I train the initialized HMMs using images of words. I used the same process for the dataset, I extract some features from the image and I transform these features to a discret sequence to be used in the HMMs parameters estimation.

I used this tool to convert the discret sequence (text) to a suitable format requested by HTK ('htkconvert' you can find it here: http://www.liacs.nl/~erwin/). You find in the joint file, an example of a sequence extracted from an image. The first number in the sequence is the length of the sequence.

I hope this can helpe you.

I already done this in HTK ,but it error and error...

I think it maybe mistake

So, I want to know step by step for doing this :)

Thank you in advance :)

Hello,

you should pay attention to everything you do with HTK (data format, HMM models ...), else you will get a lot of errors. I used HTK tool in handwriting character/word recognition using discrete HMM. Generally, there are 3 steps to use quickly HTK tool:

1- Preparing the data: it depends on the problem you working on (voice, handwriting, gene, ...). and also the type of HMM (discrete or continue).

2- Training the HMM: first, you should create HMM models (text files). then, initialisation of the models (using HTK), and finally, training the models (using HTK).

3- Test: after you trained the HMM models, now you can test and evaluate them using HTK functions.

These are general steps to use HTK, and each steps has its details. I hope this can give you a global view of how to start using HTK. It depends mainly on the problem that you treat.

I am doing online hand written character recognition, I have done preprocessing steps usng MAtlab.For the next stages i want to use HTK software.Is it possible,could  you please tell me the procedure

Would you please help in the following error that I phase in HTK