Acoustic Model Discussions

Flat
untitled
User: Visitor
Date: 7/13/2009 4:49 pm
Views: 5265
Rating: 2

Sorry, the post above is incomplete. I put it here again.

I want to use julius to recognize this two phrases:

Hello world
Security issue

This would be the grammar file:

S : NS_B SENT NS_E
SENT: WORDONE WORDTWO

 

And this de voca file:

% NS_B
<s>        sil

% NS_E
</s>        sil


 % WORDONE
HELLO    hh ax l ow
SECURITY    s ix k y uh r ax dx iy

% WORDTWO
WORLD    w er l d
ISSUE     ih sh uw

Is it possible to make an acoustic model just with this two phrases? And if it is, will this improve the recognition of this two phrases?

 

Thank you

--- (Edited on 7/13/2009 4:49 pm [GMT-0500] by Visitor) ---

Re: untitled
User: kmaclean
Date: 7/14/2009 8:51 am
Views: 179
Rating: 2

>Is it possible to make an acoustic model just with this two phrases?

Yes, but why?

>And if it is, will this improve the recognition of this two phrases?

No necessarily.  See this post for the reasons why:One word grammar, always recognized? (the later posts) which says:

It seems like Julian returns the best fit for whatever grammar you have, and if you have a one word grammar that means returning that single word every time.

[...]

 

As a workaround, you might try creating a grammar with a few Out-of-Vocabulary words, in addition to the word you want to recognize, and then use your application to determine if your target word gets recognized. 

tpvelka indicates in ths post: Whole-Word Recognition that:

Whole word HMM are usefull if you want to do IWR and do not have enough training data. Otherwise, just use phoneme HMM (i.e. the standard described in all tutorials).

Ken

--- (Edited on 7/14/2009 9:51 am [GMT-0400] by kmaclean) ---

Re: untitled
User: Visitor
Date: 7/15/2009 10:30 am
Views: 153
Rating: 2

Thank you Ken,

So I wouldn't get any improvement.

Is there any way to adapt the model as it is said in this tutorial http://www.voxforge.org/home/dev/acousticmodels/linux/adapt, but only recording those phrases.


Thak you

--- (Edited on 7/15/2009 10:30 am [GMT-0500] by Visitor) ---

Re: untitled
User: kmaclean
Date: 7/20/2009 12:07 pm
Views: 2125
Rating: 1

>Is there any way to adapt the model as it is said in this tutorial

>http://www.voxforge.org/home/dev/acousticmodels/linux/adapt, but only

>recording those phrases.

This should work (no guarantees - I have not tried adaptation with so little of speech):

This assume you have downloaded HTK v3.2.1.

Ken

 

 

 

--- (Edited on 7/20/2009 1:07 pm [GMT-0400] by kmaclean) ---

PreviousNext