VoxForge
VoxForge
Hello :-)!
I'd like to create application for mobile phone such that:
1. user enters this application,
2. user speaks ten digits, then says stop, application checks control sum, if the sum is correct it asks the user for some additional informations, if not it asks whether the user accepts improper control sum or wants to repeat digits
3. based on this talk, application creates little text file and sends it to server (probably with Tomcat)
The other possible approach (however that first one is better) is somehow different:
1. user calls special number, it connects him to server with Asterisk
2. the same as above
3. the only one difference is that there is no need to send it, it is just saved on the server.
At first I thought about CMU Sphinx, i.e. PocketSphinx (C language) for first approach or Sphinx4 (Java) for second one. However now I think that it may be better idea to check other freeware, open-source systems like HTK or Julius. Which of these three (Sphinx, HTK, Julius) is better for first, and which for the second approach and why? (I cannot buy super-expensive mobile phone so hardware resources on phone are limited).
Can you give me, please, some tutorials which say about (beginning from most important):
1. where to download the system, how to compile, install and run it on computer
2. how to create algorithm of recognition (as explained in second point above)
3. how to run it on mobile phone!
4. how to create my own model (what kind of models? acoustic and language I guess?). I need to create my own because I need other language that English and I am almost sure there won't be already existing, available for free model for my language (it is Polish language). However, maybe English model would be good enough, even if there may be some differences in phonemes.
Thank you very much for your help in advance :-)!
Greetings :-)!
--- (Edited on 10/25/2009 6:01 am [GMT-0500] by johnyjj2) ---
Hi johnyjj2,
>At first I thought about CMU Sphinx, i.e. PocketSphinx (C language) for
>first approach or Sphinx4 (Java) for second one.
>Which of these three (Sphinx, HTK, Julius) is better for first, and which
>for the second approach and why?
I think you can answer this question yourself by searching for Open Source telephony speech recognition apps and seeing what others have done... Sphinx seems to be the leader on this...
With respect to which approach (handheld-based recognition vs server-based)... it depends on your application, and how many words you need to recognize. Fewer words and you might be able to get away with hand-held based approach, with more words, then you need server based recognition.
>4. how to create my own model (what kind of models? acoustic and
>language I guess?).
see also Arthu Chan's: 10 Common Pitfalls of using SphinxTrain
Ken
--- (Edited on 10/26/2009 10:44 am [GMT-0400] by kmaclean) ---
Cross posts:
Related posts:
--- (Edited on 10/26/2009 1:41 pm [GMT-0400] by kmaclean) ---
Thanks for help!
OK, in the case of topics somewhere else I'll be giving links :-).
I followed this tutorial http://www.speech.cs.cmu.edu/sphinx/tutorial.html, there were some difficulties in Windows so I installed Ubuntu and everything was prepared. However, I couldn't execute examplary code, as I explained here: https://sourceforge.net/projects/cmusphinx/forums/forum/5471/topic/3445960
At this moment the most important need is to run it on computer and later on mobile, example with English language.
This http://www.voip-info.org/wiki/view/Sphinx can wait for later. And about this http://www.keithv.com/software/parakeet/ I guess I need some more tips, I don't need the feature of touch-screen and I don't know how to enable using pocket sphinx on mobile with parakeet.
Later, I would create those models. I downloaded files contained in Germany model and tried to understand what is where and what for. My first impression is that it contains so much redundant data. Many things are repeated in many files in similar ways.
But at this moment it is about the simplest thing, i.e. running sphinx3/4 on laptop and pocket sphinx on mobile. I guess with proper tutorials to read and also working examples, I would be able to try to do something analogically to already existing things.
Thanks once more for your help :-)!
--- (Edited on 10/31/2009 11:53 am [GMT-0500] by johnyjj2) ---