I'm new using HTK. I want to use the HMM for non speech data and I haven't found any good example, mostly all examples are for speech.
I know that I have to convert the feature of my data to the HTK format.
What I don't know, is how to configure HTK, because it uses a dictionary and a grammar, and I won't be needing those things with my data because its not speech.
Does someone has a good example or any advice for this?
I recently worked on HTK-based OCR and found this ICFHR tutorial is very helpful (http://transcriptorium.eu/~tutorialICFHR/).
The only essential difference is your feature, and thus it will be sufficient to refer the HTK book for all configurations. Of course, you need some modifications to some of configurations and HMM definitions, but the idea of GMM-HMM never changes. For example, here is a mapping table between ASR and OCR.
ASR: phone | tri-phone | lexicon [ word : decomposed_phone_sequence ]
OCR: char | tri_char | lexicon [ word : decomposed_char_sequence ]
My suggestion is to read the HTK book through-fully and to play with some demo codes. You probably will get confused when you first work with HTK, especially for its command line params, but you will feel comfortable in a month.
