It working excellent with Simulator. However it not working fine with Ipad
device. getting more time to display the detected voice also it has only 30%
Accuracy where simulator had 95-100 % accuracy.
i am unable to fin out the issue. may i know any valuable suggestions for your
side
Thanks
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
voxforge-en-0.4 is 16khz model for desktop microphones, it's not compatible
with iphone audio which is 8khz. You need to use 8khz acoustic model like
hub4wsj_sc_8k provided with the pocketsphinx.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Let me clarify one thing,
Would it be possible to obtain more accuracy than openears default acoustic
model using the custom trained one with first 20 sentences from CMU ARCTIC
text-to-speech databases which is described in http://cmusphinx.sourceforge.n
et/wiki/tutorialadapt
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
You need to understand that accuracy improvement is a systematic process. It
requires you to run a specific set of actions to get a result. You can read
more here
As for your question, please avoid asking for confirmation of the obvious
things or things that were already written before. It's not a productive way
of communication.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
I run both voxforge-en-0.4 and hub4wsj_sc_8k on iPhone 4 with pocketsphnx 0.61
live with OpenEars and the 16k models are more accurate. I also run0.61 and
0.7 pocketsphinx_batch (an interesting experience on an iDevice) and see
exactly the same results running live, voxforge has about 40% lower WER than
WSJ.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
i wish to have a clarification in acoustic model implementation
I have done the below process:
1.Downloaded voxforge-en-r0_1_3.tar.gz and extracted.
2.copied voxforge_en_sphinx.cd_cont_3000 folder and pasted it in my Openears
Library Folder(/Users/mac01/dev/OpenEars/CMULibraries/pocketsphinx-0.6.1/model
/hmm/en_US)
3.Renamed the folder as hub4wsj_sc_8k
4.Cleaned all targets and Built Openears Project file.
may i know is there any more steps to do?
Thanks
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Hi all
i used the English Voxforge Accoustic model from http://sourceforge.net/proje
cts/cmusphinx/files/Acoustic%20and%20Language%20Models/ with my
Openears.
It working excellent with Simulator. However it not working fine with Ipad
device. getting more time to display the detected voice also it has only 30%
Accuracy where simulator had 95-100 % accuracy.
i am unable to fin out the issue. may i know any valuable suggestions for your
side
Thanks
voxforge-en-0.4 is 16khz model for desktop microphones, it's not compatible
with iphone audio which is 8khz. You need to use 8khz acoustic model like
hub4wsj_sc_8k provided with the pocketsphinx.
So may i know which will be the better Option for me to work with Ipad.
is US **English Communicator Telephone Acoustic ** will work?
i have downloaded 8KHZ English Voxforge Accoustic model from
http://www.voxforge.org/home/news/news/updated-8khz-sphinx-acoustic-
model
However i found the same issue..
Working fine with Simulator but not well with IPad.
corrected link is http://www.voxforge.org/home/news/news/updated-8khz-sphinx-
acoustic-model
I told you above: hub4wsj_sc_8k provided with pocketsphinx distribution.
hub4wsj_sc_8k provided with pocketsphinx distribution is not getting better
accuracy. may i know is any other better acoustic models available?
No, hub4wsj_sc_8k is the best one.
Let me clarify one thing,
Would it be possible to obtain more accuracy than openears default acoustic
model using the custom trained one with first 20 sentences from CMU ARCTIC
text-to-speech databases which is described in http://cmusphinx.sourceforge.n
et/wiki/tutorialadapt
You need to understand that accuracy improvement is a systematic process. It
requires you to run a specific set of actions to get a result. You can read
more here
http://stackoverflow.com/questions/7430337/very-low-accuracy-while-using-
open-ears-for-speech-
recognition/7443545#7443545
As for your question, please avoid asking for confirmation of the obvious
things or things that were already written before. It's not a productive way
of communication.
Thanks for your professional and effective help.
I run at 16k, no problems. An iPhone 4 will quite happily sample at 44.1k. I
believe a 3 will, too.
Mine is running fine with 16k models.
I run both voxforge-en-0.4 and hub4wsj_sc_8k on iPhone 4 with pocketsphnx 0.61
live with OpenEars and the 16k models are more accurate. I also run0.61 and
0.7 pocketsphinx_batch (an interesting experience on an iDevice) and see
exactly the same results running live, voxforge has about 40% lower WER than
WSJ.
Thanks, wiz, it should be so.
In any case it needs extensive testing beforehand. Trying different models is
the best way to waste a time.
@wiz:
i wish to have a clarification in acoustic model implementation
I have done the below process:
2.copied voxforge_en_sphinx.cd_cont_3000 folder and pasted it in my Openears
Library Folder(/Users/mac01/dev/OpenEars/CMULibraries/pocketsphinx-0.6.1/model
/hmm/en_US)
3.Renamed the folder as hub4wsj_sc_8k
4.Cleaned all targets and Built Openears Project file.
may i know is there any more steps to do?
Thanks
@wiz: @ nshmyrev: Also please verify the links that i downloaded.
**16 KHZ ** downloaded from http://sourceforge.net/projects/cmusphinx/files/Acoustic%20and%20Language%20Models/
8 KHZ downloaded from http://www.voxforge.org/home/news/news/updated-8khz-sphinx-acoustic-model
waiting for your valuable suggestions..
The Openears Details:
OpenEars Version 0.913
pocketsphinx-0.6.1
sphinxbase-0.6.1
no
links are correct