Any hints for Arabic user custom traineddata (e.g. new font)

[bmwmine]

Hi everybody

I am creating curpus data to train. It is more than 9000pages (like 3 encyclopedias) of tif for "Traditional Arabic" font (will add other fonts later)

will use Tesseract 4.00 Alpha LSTM any hints will be useful.

I took same config of recent ara.traineddata (12mb). set char_spacing to 0.7

will use this lstm paras

--learning_rate 10e-5
--net_spec '[1,0,0,1 Ct5,5,16 Mp3,3 Lfys64 Lfx128 Lrx128 Lfx256 O1c1]'
--net_mode 192
--perfect_sample_delay 19

as referenced

Also, It is good to share your experience with tesseract ocr engine if anybody has custom trained data for Arabic please share it