ARPA or JSGF?
ARPA. I have checked the file content, it is supposed to be alphabetically sorted. However, due to special characters in Turkish (ç, ğ, ş etc.), it is not actually sorted in Turksih. This may have some effect, but ARPA I used on Linux machine and iPhone project is same.
Is the acoustic model 8-bit or 16-bit?
Acoustic model is created with wav files that are 8kHz, 16bit, mono.
What is the size of the language model or grammar?
In acoustic model the number of words in dictionary is 400+. To get a good accuracy I created a language model and dictionary with 7 words, which are days.
How far away from the phone are you when speaking? (this has a huge effect for dictation applications)
I speak directly into the mic.
What are you trying to recognize? Words, sentences, phrases?
As I mentioned before it is just the days, 7 words.
Can I ask you for some clarification about this question?
I thought maybe the format of iOS takes speech and feeds it to pocket_sphinx have effect in performance.
Here is a sample log: http://ug.bcc.bilkent.edu.tr/~m_kalender/Turkish_log.txt
I see you point, but still the big performance difference makes me search something in iOS to find out the reason.