Hi erica,

Are you using Windows or Linux?  Did you update your user path after you compiled HTK?

Can you go into the directory where you compiled HTK and execute HDMan from there? 

Note that in Linux (and I think Cygwin is the same ...) commands are case sensitive.  That means that "HDMAN" is not the same command as "HDMan" (you want HDMan).

Hope that helps,

Ken 

Hi,

       thanks for your help, I succeded in using the HDMan command.

J have just an other question. Using my .wav file I build up a speaker dependent acustic model. But what should I do to have a sepaker indipendent acoustic model?My problem is that I'm trying to use julian to recognize only personal name (simple grammar) but with different voices. Do you think that for each voice will I have to insert a .wav specific file? Do you think that julian is a good choiche for this?

Thanks,  

 

 

>But what should I do to have a sepaker indipendent acoustic model?

I don't know ... it depends on your application.  The acoustic models used with  the Sphinx group of speech recognition engines use 140 hours of speech.  It gets reasonably good results recognizing speech from people who have not contributed their voice to the creation of the acoustic model.

>My problem is that I'm trying to use julian to recognize only personal name

>(simple grammar) but with different voices.

This makes things easier since you don't have as large a grammar to deal with.  So you likely won't need 140 hours, but I don't really know exactly how much you would need. 

>Do you think that for each voice will I have to insert a .wav specific file? 

I would *guess* that 5 minutes of audio from 30-50 people might give you a good start, recording your simple grammar  (less than 30 known names).

In addition, there are other things you can do to improve performance greatly with a small grammar recognition system:

• using a noise model,
• using word-based hmm with more states (rather than phone-based hmm),
• not using context-independent models

that are not discussed in the Voxforge howto/tutorial.  Regardless, getting the audio is a good first step.

>Do you think that julian is a good choiche for this?

Well if you have the audio, and you realize that Julian does not give you the results you are looking for, you can always try HTK or create an acoustic model for Sphinx.

Ken 

please tell me how you solved the problem with the HDMan command!