Acoustic Model Discussions

Flat
Continuous speech dictation with lm_giga_64k_nvp_3gram
User: mendroid
Date: 9/29/2009 4:51 am
Views: 6570
Rating: 6

Hi,

i downloaded the file: lm_giga_64k_nvp_3gram and converted  to .bin format ( julius format ) with the command:

mkbingram -nlr lm_giga_64k_nvp_3gram.arpa julius.bin

but now i need to know how use it.

1.- I need to define the files .voca and .grammar ? In this case where can i find these files for the new .bin file?

2.- How "exactly" i need configure the file julian.conf

3.- I need more files to try continuous speech dictation with Julius?

 

Thanks in advance,

Mendru

--- (Edited on 9/29/2009 4:52 am [GMT-0500] by mendroid) ---

Re: Continuous speech dictation with lm_giga_64k_nvp_3gram
User: kmaclean
Date: 10/2/2009 9:08 am
Views: 2655
Rating: 5

Hi Mendru,

>1.- I need to define the files .voca and .grammar ?

You don't need these for continuous dictation - the language model replaces the grammar.

>2.- How "exactly" i need configure the file julian.conf

Have you looked at the Julius Book?

Based on this post, it seems like you need to following settings:

  • language model

    • -d julius.bin

    • -v dictionary used to create lm_giga_64k_nvp_3gram

  • acoustic model
    • -h hmm15/hmmdefs

    • -hlist tiedlist

>3.- I need more files to try continuous speech dictation with Julius?

you need an acoustic model, in addition to the language model you downloaded.

Ken

--- (Edited on 10/2/2009 10:08 am [GMT-0400] by kmaclean) ---

PreviousNext