Adapting Acoustic Model with Julius

German

User: Nikita
Date: 1/29/2013 8:14 am

Views: 6500
Rating: 5

Dear VoxForgians!

I would like to train the most recent German AM for HTK/Julius according to the tutorial (http://www.voxforge.org/home/dev/acousticmodels/linux/adapt/htkjulius). It requires the following files:

hmmdefs
macros
tiedlist
config
wav_config
dict
stats

of which I don't have "config", "wav_config" and "dict". I figured that "dict" contains words and corresponding pronunciation. But does it have to be in that exact format of:

Word [WORD] w o r d ? Or can I move on with only the first and last culumns?

All hints will be very much appreciated :)

Best wishes

Nikita

P.S.: Has anyone tried adapting AMs with HTK 3.4.1? What were your experiences?

Re: Adapting Acoustic Model with Julius

User: nsh
Date: 1/31/2013 11:00 am

Views: 142
Rating: 5

> Or can I move on with only the first and last culumns?

Yes, you can

> Idon't have "config", "wav_config"

You can find them in voxforge tutorial

http://www.voxforge.org/home/dev/acousticmodels/linux/create/htkjulius/tutorial/data-prep/step-5

> Has anyone tried adapting AMs with HTK 3.4.1? What were your experiences?

CMUSphinx is easier to deal with, see

http://cmusphinx.sourceforge.net/wiki/tutorialadapt

Re: Adapting Acoustic Model with Julius

User: Nikita
Date: 2/3/2013 9:54 am

Views: 2874
Rating: 5

Hello Nickolay!

Thank you very much for the answers!

Previous • Next •


Username	Password