A Transkribus-Trained ML Tool for Historical Italian Manuscripts

In this project, I aim to develop a trained machine learning model to transcribe historical Italian texts (1300s-1700s). I examine the efficacy of contemporary OCR and Convolutional Recurrent Neural Network (CRNN) techniques, and ultimately introduce a model based upon the Transkribus Print M1 Model. My model is trained on a dataset of transcriptions from the Newberry Library (Chicago, IL), and exhibits a CER rate of 10.30%.