Hi Manuel,

The phone list used in Festival is here.

The Italian dictionary is here.

Ken 

Perfect. You're unique!!!!

Now I try to create my italian language model.

Someone can give some advice about how many audio record to train the model?

Tks

Manuel 

... after many hours I understand that my model can't be create if I don't modify the question of tree.hed, but I have no idea how to do it.

Now I try to create english acoustic model, but always I have very, very, very high Word Error Rate, but I don't understand why. I train the model with over than 100 audio files...

Bye

Manuel 

First of all, why not use ready sphinx setup from voxforge. Sphinx can build tree automatically.

Second, if you have good book on italian phonetics, i can build a questions for you.

Third, just share your files, we'll try to help.

Hi nsh,

Can Sphinx automatically build questions that can be used with HTK?

Ken 

 

sure, but handmade questions are better. Current questions are in model_architecture subdir in svn.

I've just wWrote a little text about questions, comments and improvements are welcome:

http://www.dev.voxforge.org/projects/Main/wiki/AcousticTreeQuestions

Hi,

I tried to access the linked Italian phone list and dictionary, but a password is required.

What should I do to access these documents?

 

Thank you and bye

Mimmo

Hi Mimmo,

Try here (using Trac rather than SVN):

The phone list used in Festival is here.

The Italian dictionary is here.

Ken

Thank you Ken for your fast answer.

I am evaluating Sphinx in order to recognize a small set of pre-recorded phrases, as  soon as I reach a stable prototype I will post a description of the work.

Bye

Mimmo