Add Sámi tesseract model

smi/README.md

@@ -0,0 +1,20 @@
+# Smi: a tesseract model for Sámi
+
+This model was developed for OCR on the Sámi printed material in the collection of the National Library of Norway.  
+
+It can be used for OCR on the following Sámi languages:  
+- North Sámi
+- South Sámi
+- Lule Sámi
+- Inari Sámi 
+
+
+## Training
+This model has been trained from the [Norwegian tesseract model](https://github.com/tesseract-ocr/tessdata_best/blob/main/nor.traineddata) from tessdata_best  
+It has first been trained for 5 epochs on [Sprakbanken/synthetic_sami_ocr_data](https://huggingface.co/datasets/Sprakbanken/synthetic_sami_ocr_data).  
+Then, it has been fine-tuned on a dataset consisting of 93 415 lines of Sámi and Norwegian OCR data (a combination of manually annotated and automatically transcribed data).  
+We unfortunately cannot share the data as it copyrighted material.  
+
+
+Read more about the model training and evaluation in the paper "Comparative analysis of optical character recognition for Sámi texts from the National Library of Norway", accepted for NoDaLiDa/Baltic-HLT 2025 ([link to preprint](https://arxiv.org/abs/2501.07300))
+