VoxForge
VoxForge
hi all,
im going to develop a speech recognition system using my native language for dialing a number and i need the recognized sinhala speech as an english text output like below.
call 0 1 1 3 4 5 6 3 2 6
my native language is sinhala and for an example ,
call = "amathanna"
1 = "eka"
2 = "deka" etc
so my speech will be "amathanna eka deka thuna... " like that.
so can you please tell me how im going to start this project using htk + julian??
Is it possible to do by using julian?
And also tel me how to write the task grammer for my problem.. Thanx a lot.
--- (Edited on 3/22/2011 11:20 pm [GMT-0500] by ) ---
>so can you please tell me how im going to start this project using htk + julian??
First, get a better understanding of how to do this in English using the VoxForge Tutorial.
Then create a pronunciation dictionary for sinhala and record some speech (using words from your pronunciation dictaionary), and then create a monophone acoustic model (up to step 8 in the VoxForge Tutorial).
For better recognition results, you can thereafter create a triphone acoustic model, but this requires a sinhala "question" file.
The problem with HTK/Julius is thht you need to create the "question" file yourself. Whereas the CMU Sphinx acoustic model trainer: Sphinxtrain can create one for you. Manually created "question" file are usually better, but you may need some linguistic training to create one yourself.
--- (Edited on 3/30/2011 10:42 am [GMT-0400] by kmaclean) ---