Some old texts would probably have been OCRable when they were new, but t̷h̵e̴ y̴̿ͅe̶͙̾̔ȃ̸͙̝͌r̷͔̈́s̷͉͙̈ r̵͉̣͔͈̘̈́̽̈́̀u̶̧̧͓̼̹̞͉͎͆̇͊̍̀̄i̶͉͎̟̘̙͈̊̀̈́̈͗̆̅n̴̨̗̰̮̫̐ͅ m̵̧̡̨̫͉̥͓͚͑͊͌̈́͜a̴͎͚̠͚͓͈͛̈́̇̍́̚͝ṉ̶̛͍̻͖̬̿̔̓͜y̵͖͙̮̘̮̞̑̓̀̉̚ ̷̨̢̝̣̰͇͌t̶̖͙͂̇̈̌͆̈́̀̈́̓h̵̦̜͍̲̽͐̔̆̅̓͌i̷̡̮͍͈̎̐̿̉n̶̩̺͎̹͋͒̇̒̍̌̉͝g̸͔̓̒s̴̢̖̯̞̗͇̳̦̀͆̔̈́͑͌̕͝
Comment on TIL (Trauma I Learned)
ivanafterall@kbin.social 1 year ago
I'm writing a book. I have nearly 500 old newspaper sources of wildly varying quality and formats, as well as journal entries and the like. You'd think it'd be as simple as OCR, but OOOOHHHH NOOOO! I'm so happy someone understands my nightmare.
starman2112@sh.itjust.works 1 year ago
fossilesque@mander.xyz 1 year ago
readcoop.eu/transkribus/ <- Worth a shot.
Also:
ivanafterall@kbin.social 1 year ago
Curious whether you had luck with any of these? I found Transcribus awhile back and heard angels singing. It seemed so perfect. But I didn't have much better luck than with OCR options out of the box and custom training a model seems like nearly as much work as just transcribing, and I'm skeptical the result wouldn't be flawed and require a lot of manual fixes.
fossilesque@mander.xyz 1 year ago
I have had pretty good luck with it, but also think about considering trying ChatGPT now that it’s got vision. There are now libraries for typed print on transcribus. Worth a shot, since you get 500 free credits.