VoxForge
VoxForge
Hi All,
I have Julius up and running on my Linux system and have been able to understand and add to the .grammar and .voca files. My question is this:
I want to use Julius for dictation. I want to be able to talk at Julius in english as naturally as possible and have it ouput what I am saying as best it can. Has anyone created generic .grammar and .voca files for this that include all the words in the VoxForge Accoustic Models?
What is the current best way to set up Julius for dictation?
--- (Edited on 12/30/2008 11:36 am [GMT-0600] by ) ---
This trigram model is generic enough:
http://www.inference.phy.cam.ac.uk/kv227/lm_giga/lm_giga_64k_nvp_3gram.zip
You can use it with julius with -d option when you convert it to a juluis binary format with mkbingram.
That will not work well though.
--- (Edited on 12/30/2008 4:52 pm [GMT-0600] by nsh) ---
Hi slidgey,
>So how do I get dictation working?
>Julius claims to work well in Japanese. Why not in english
Because we need way more *English* speech (1000+ hours) to create acoustic models that might be used for dictation.
Ken
--- (Edited on 1/2/2009 11:03 am [GMT-0500] by kmaclean) ---
What command line are you using with mkbingram? I keep getting an "invalid header" error.. sure it's a newbie mistake but I'm trying to get something at least semi-usable I can work with. I realize the model isn't yet to the point where it can handle dictation, but without a larger grammar file it's hard to get over the initial learning curve.. thanks for any advice you can give.
--- (Edited on 5/5/2009 11:34 pm [GMT-0500] by Visitor) ---
Hmm.. I get the help text with those options:
$ ../bin/mkbingram -nlr lm_giga_64k_nvp_3gram.arpa julius.bin
mkbingram: make/convert binary n-gram from ARPA format for Julius
usage 1: ../bin/mkbingram ARPA_2-gram ARPA_rev-3-gram outfile
usage 2: ../bin/mkbingram -d bingram outfile
--- (Edited on 5/8/2009 5:32 pm [GMT-0500] by Visitor) ---