Why make the recognition with julius and not  with the tool htk  "HVite "

 I followed the stages of htk book and I made the recognition with Hvite but with a model monophonem and I have a rate of recognition of 38%
Is it logical??

Note: I made a training with the data base timit ( great corpus of data).

--- (Edited on 9/10/2008 7:58 am [GMT-0500] by imene) ---

--- (Edited on 9/10/2008 8:22 am [GMT-0500] by imene) ---

Hi imene,

I'm not sure I understand everything you are asking, but here goes...

>Why make the recognition with julius and not  with the tool htk  "HVite "

Because there was no tutorial for creating an acoustic model for Julius, and HTK has its own tutorial.

You can still use HVite with the acoustic models you create in the VoxForge tutorial... Just use the same instructions from the HTK book, and modify your config file to have the same feature set as the VoxForge acoustic model (look at the config file in the nightly builds; though you will probably get better recognition with the current release)

> I followed the stages of htk book and I made the recognition with Hvite

>but with a model monophonem and I have a rate of recognition of 38%

>Is it logical??

You tell me... :) 

It's been a while since I looked at HVite, but I seem to recall getting better recognition with Julius than with HVite, even with the same acoustic model.

>Note: I made a training with the data base timit ( great corpus of data).

The VoxForge corpus will some day be better!

Ken

--- (Edited on 9/10/2008 10:35 am [GMT-0400] by kmaclean) ---