While trying to fix some failures of IA Upload an unexpected result emerged: an easy opportunity of fixing some usual OCR errors into djvu text layer.
In brief, the script
xml2dsed.py converts IA _djvu.xml files into a "dsed" (lisp-like) code, so that text layer can be uploaded into djvu file into a much faster and controllable way using djvused.exe. While parsing the xml tree, at WORD level any word of the text layer is exposed to the script environment as pure text; this offers a unique opportunity to fix many scannos, avoiding any risk to mess the xml or the dsed code.
Alex brollo
