How to generate pdf sequence from realigned model

Speech Recognition Engines

Flat

User: Visitor
Date: 6/14/2010 12:50 am

Views: 4850
Rating: 5

Hi,

I am training acoustic model and reached Step 8. I have aligned model and align dat is like :-

#!MLF!#
 "xxxxxxxxx.lab"
 0 400000 sil[2] 1058.880981 sil
 400000 3250000 sil[3] 7959.713867
 3250000 3300000 sil[4] 132.230301
 3300000 3650000 sil[2] 934.297424 sil
 3650000 3700000 sil[3] 120.060242
 3700000 4500000 sil[4] 1768.828125
 4500000 5850000 ao[2] 3013.056641 ao

Now I want to generate pdf sequence for synthesis, but I am not able to get head-start.

My frame rate is 5ms and 5 states. I understand that first 2 columns are start and end time and label indicates state.

Can you please tell how to generate pdf out of this data.

Thanks,

AD

--- (Edited on 6/14/2010 12:50 am [GMT-0500] by Visitor) ---

Re: How to generate pdf sequence from realigned model

User: kmaclean
Date: 6/14/2010 8:09 am

Views: 81
Rating: 5

>I am training acoustic model and reached Step 8.

The output you are shoing is not from Step 8 of the VoxForge Tutorial ... what is you HVite command line?

>I want to generate pdf sequence for synthesis,

I am not familiar with this... best place to ask would likely be the HTK mailing list.

Ken

--- (Edited on 6/14/2010 9:09 am [GMT-0400] by kmaclean) ---

Re: How to generate pdf sequence from realigned model

User: AD
Date: 6/15/2010 9:52 am

Views: 2111
Rating: 5

HVite -A -T 1 -a -f -m -C config -H MMF -I MLF -i AlignmentOutFile -S SCPFile Dictionary ModelList

I used -f option for state level alignment, thus state duration.

Thanks, I will pursue this in HTK forums.

Regards,

--- (Edited on 6/15/2010 9:52 am [GMT-0500] by AD) ---

Previous • Next •


Username	Password