VoxForge
VoxForge
usually my forum posts are a little short tempered due to my frusteration with trying and researching and trying something and posting as a last resort so please bear with me...
I'm using Julius-4.1.4 with Julian enabled.
I understand that there is an acoustic model that contains statistical data of voice recordings. Where is the best free US english model I can get?
I understand there is some sort of dictionary that lists words and phonemes. Where is the largest free US english dictionary I can get? If it is not in Julius format then how to do I convert? (i've found some at cmu)
Does Julius need dictionary, grammer, voca, dfa files or only Julian?
How do you match up your grammer files with your acoustic model?
All i've seen in action so far is the sample files included with the Julian quickstart. I am impressed with its recognition, however I would now like to take it further. The voice commands I would be using would be for home automation so they would mostly be canned phrases such as "Computer, turn the light on", however there could be an occational random utterance such as "Look up xyz" (on the internet)
Please be gentle, no short know-it-all answers or RTFM answers because I have dont lots of reading...
--- (Edited on 2/3/2010 6:06 pm [GMT-0600] by pomprocker) ---
>I'm using Julius-4.1.4 with Julian enabled.
Julian does not exists in Julius 4.1.4 - Julius does both grammar recognition and recognition using statistical language models.
>I understand that there is an acoustic model that contains statistical data
>of voice recordings. Where is the best free US english model I can get?
The Sphinx acoustic models are good. Keith Vertanen has created good acoustic models for HTK and Sphinx (Google it...)
>I understand there is some sort of dictionary that lists words and
>phonemes. Where is the largest free US english dictionary I can get? If it is
>not in Julius format then how to do I convert? (i've found some at cmu)
CMU is good. There is an HTK tools to do this (I think it is HDMan, but can't remember for certain...), or create your own script (Perl, Bash). The Julius manual 3.2, section 4 Word dictionary files describes the required format.
Acoustic models are closely tied to the pronunciation dictionary used to create it.
>Does Julius need dictionary, grammer, voca, dfa files or only Julian?
Julian no longer exists in version 4.x of the Julius speech recognition engine.
Look at the VoxForge Tutorial Running Julian Live (which uses Julius 3.2). The config files and required files are very similar, and use 'julius' rather than 'julian'.
>How do you match up your grammer files with your acoustic model?
Use the pronunciations contained in the pronunciation dictionary that was used to create the acoustic model.
>All i've seen in action so far is the sample files included with the Julian quickstart.
See Step 1 of the VoxForge Tutorial to see how to expand your grammar files.
Ken
--- (Edited on 2/3/2010 8:56 pm [GMT-0500] by kmaclean) ---
I really don't want to create my own files if I don't have to. Aren't there pretty decent ones already out there that I can use out of the box to recognize a large assortment of words?
--- (Edited on 2/3/2010 8:17 pm [GMT-0600] by Visitor) ---