1. Why train a Text Recognition Model?

Public models are great for general use and quick results, while custom models provide the precision and adaptability needed for specialised projects.

Transkribus offers public, pre-trained models for general use across different languages, document types and handwriting styles. However, if you're working with unique handwriting or specialised documents, a custom model can significantly improve transcription accuracy. This article explains the differences and benefits of using a custom model.

1. Using a Public Models

Public models are pre-trained on a wide range of documents, making them an excellent choice for general transcription tasks. 

• Ready to Use: Public models are pre-trained and can be applied immediately to your documents.
• Large & Robust: They are trained on extensive data, making them effective across a wide variety of text types.
• Easy to use: Designed to be accessible for new users and straightforward to use without any setup.
• Versatile for Handwritten and Printed Text: Many public models are built to handle both handwriting and printed text, suitable for general purposes.

When to Use: Public models are ideal if you’re working with common handwriting styles, printed materials, or if you want a quick start without needing specialised accuracy. Visit this article to find out more about how to use public models: Automatically transcribing documents.

2. Training a Custom Model

Custom models are tailored to your specific documents, allowing you to achieve higher accuracy and more consistent results for unique or complex materials. Here’s what custom models offer:

• Higher Accuracy: Because custom models are trained on your own documents, they learn your specific handwriting styles and document layouts, resulting in more precise transcriptions.
• Adaptable to Your Material: Custom models can handle unique vocabulary, terminology, or layouts that might be challenging for public models.
• Control Over Training Data: You decide what data goes into the training, allowing you to shape the model around the exact types of documents you’re working with.
• Fine-Tuned for Your Project: Custom models can be continually improved with additional data, making them perfect for specialised, historical, or multi-author documents.

When to Use: Custom models are ideal for specialised projects, unique handwriting styles, or when public models aren’t delivering the accuracy you need. They’re especially useful for any task that requires consistent, high-quality transcription. To learn more about training a custom text recognition model, go to the next step on Training Text Recognition Models.