Is there a way to prepare ".box" files using your own "traineddata"?

[Rafael Maciel K]

I've been training tesseract to recognize well a new font. I already generate a new "traineddata" for this font, and now i want to add more data to my training set.

The problem is that when I generate the ".box" files, looks like its using the "eng.traineddata" to make the recognition. How can I generate them using my own ".traineddata" (since at this point my ".traineddata" is more accurate for this font, it would generate less rework)?