VoxForge
VoxForge
rks_mfs asks:
Hi,
I am trying to get Sphinx to work with Indian languages and thinking of working with phone set for one language at a time.
What is the best approach to do this in Sphinx. Can I directly work with phoneme decoder so that I do not have to get into Vocabulary and word dictionary.
Does Sphinx do phoneme only output and is the recognition real time.
for instance if I say "Hello" how can i get a phoneme which recognize 'h' first then 'e' then 'l' then 'l' then 'o' each independently?
Thanks.
--- (Edited on 6/18/2008 12:13 pm [GMT-0400] by kmaclean) ---
Hi rks_mfs,
>I am trying to get Sphinx to work with Indian sub-languages and thinking of
>working with phone set for one language at a time.
Do you mean you want Sphinx to recognize speech from more than one Indian language at the same time? Why? what is the application you are thinking of?
>thinking of working with phone set for one language at a time.
Do you mean that you want to have a separate phone set for each Indian sub-language?
>for instance if I say "Hello" how can i get a phoneme which recognize 'h'
>first then 'e' then 'l' then 'l' then 'o' each independently?
Why would you want to do this? What are you trying to do?
Strictly speaking, speech recognition works with phones rather than phonemes, so that it only distinguishes between distinct units of sound. If in your target language, a letter matches directly to a sound, then you should be able to use speech recognition to distinguish the letters. This is *not* the case in English.
Hope that helps,
Ken
--- (Edited on 6/18/2008 12:28 pm [GMT-0400] by kmaclean) ---
--- (Edited on 6/19/2008 3:48 am [GMT-0500] by Visitor) ---
Hi rks_mfs,
>For grapheme-based acoustic modeling, where can I find the steps necessary
>to build the dictionary by treating each character as a phone.
Use Google.
This first article speaks directly to your question:
These articles talk about using English acoustic models to bootstrap phone models for a new language:
Ken
--- (Edited on 6/19/2008 8:13 am [GMT-0400] by kmaclean) ---
Btw, Hindi models for sphinx were recently released:
http://sourceforge.net/projects/hindiasr/
--- (Edited on 6/19/2008 3:25 pm [GMT-0500] by nsh) ---
--- (Edited on 6/25/2008 12:02 am [GMT-0500] by Visitor) ---
Hi,
>So can you please guide me what are the steps we will have to follow to do
>this work.
See the Sphinx Manual: Hieroglyphs
Ken
--- (Edited on 6/25/2008 10:13 am [GMT-0400] by kmaclean) ---
You might also try the CMU Robust Group Tutorial (dated Aug 2007).
--- (Edited on 7/21/2008 2:45 pm [GMT-0400] by kmaclean) ---