Slightly off-topic but possibly also of interest.  In my limited testing  of SanskritCR,   it seems to be better with the fonts of the  printed editions from the late 19th and early 20th century than with newer computer fonts
Harry Spier


On Fri, Jun 17, 2022 at 2:18 PM Dominik Wujastyk <wujastyk@gmail.com> wrote:
The https://aksharamukha.appspot.com/converter service can take a multi-page PDF or image file.  But it does a less good job than https://sri.auroville.org/projects/sanskrit-ocr/

On Thu, 9 Jun 2022 at 17:16, Oliver Hellwig via INDOLOGY <indology@list.indology.info> wrote:
Thanks a lot, Harry, for explaining the technical background. This
definitely explains the strange encoding. For the time being, re-OCRing
it seems to be one manageable approach (already pointed out by Tim
Cahill), so I will try the Google cloud SDK which works in the
background of SanskritCR.

Best, Oliver