nikk6174/historical-spanish-ocr

🔍 Historical Spanish OCR Pipeline

An interactive OCR evaluation tool for 17th-century Spanish manuscripts.

Features

• —TrOCR fine-tuned on historical Spanish text (5–7% CER)
• —Rule-based correction with archaic Spanish dictionary
• —T5 post-correction trained on real TrOCR errors
• —Dynamic ground-truth alignment (Needleman-Wunsch)
• —Accent-normalised CER metric

Usage

1. 1.Upload a full-page document image
2. 2.Upload the ground-truth transcription (.txt, one line per text line)
3. 3.Click "Run Analysis"

Models are automatically downloaded from Hugging Face on first use.