HTR Software: Comparisons & Alternatives
Choosing a transcription tool means weighing accuracy, setup effort, and cost. This hub compares the main options — specialist HTR platforms, general OCR, and general-purpose AI models — and explains where each fits, and where a zero-shot model like Leo's changes the trade-offs.
12 articles
Leo Team · August 10, 2026
Crowdsourced Transcription: What Volunteers Do Best, and Where HTR Should Take the First Pass
What crowdsourced transcription volunteers reliably deliver, what programme figures do not prove, and how to decide series by series when HTR should take the first pass.
Read the article
Leo Team · August 10, 2026
Do You Need to Train AI to Read Handwriting? A Decision Guide Before You Spend a Term on Ground Truth
Whether historians need to train a custom handwriting recognizer, what ground-truth labour actually costs, and the narrow conditions under which training beats testing a ready model first.
Leo Team · August 10, 2026
Zotero for Primary Sources: Where Transcribed Text Fits in a Citation-First Workflow
Zotero for primary sources as a citation-first workflow: how to map archival fields, where checked transcriptions should live, and how to keep them anchored to leaves.
Leo Team · August 4, 2026
Amazon Textract Handwriting Recognition on Historical Records: What It Reads, and Where the Pipeline Breaks
Amazon Textract’s documented handwriting support versus its quiet failure modes on historical manuscript series, and how archives should route records and validate coverage.
Leo Team · July 26, 2026
Can Gemini Read Old Handwriting? What Happens When You Ask It to Transcribe a Manuscript
Tests whether Google's Gemini chat model can accurately transcribe historical handwriting (secretary hand, Kurrent, notarial cursive), arguing it silently normalizes and confabulates text rather than flagging errors, unlike specialist HTR tools.
Leo Team · July 26, 2026
OCR4all for Handwriting: Why a Print-OCR Toolkit Is the Wrong Starting Point for Manuscripts
Explains why OCR4all, a print-OCR pipeline, is unsuited to handwritten manuscripts and points readers toward specialist HTR tools, including Leo's zero-shot Latin-script model, as the correct alternative.
Leo Team · July 26, 2026
Omeka Transcription: How to Make an Omeka S Collection Genuinely Searchable
How to make an Omeka S collection's transcribed text actually searchable, covering the core-search/Solr/IIIF indexing chain, transcription source options (Scripto, DataScribe, OCR/HTR, Leo), and why preserving source fidelity at the transcription step matters most.
Leo Team · July 26, 2026
Leo vs eScriptorium: A No-Training eScriptorium Alternative for Reading Latin-Script Manuscripts
Compares Leo to eScriptorium/Kraken for transcribing Latin-script manuscripts, arguing eScriptorium requires finding or training a recognition model first while Leo offers ready zero-shot transcription, with honest tradeoffs on model coverage, infrastructure cost, and Leo's own limitations.
Leo Team · July 26, 2026
Adding Transcription to Tropy: Reading the Photos You've Already Organized
Explains that Tropy manages and describes research photos but does not transcribe them, and walks through adding an HTR step (comparing OCR, general LLMs, specialist HTR, and Leo) to a Tropy-organized collection.
Leo Team · July 22, 2026
The Best Handwriting Transcription Software for Historical Documents: An Honest Comparison
Compares handwriting transcription tool classes for historical manuscripts — general OCR, LLMs, crowdsourcing, and specialist HTR (trained vs. zero-shot) — arguing specialist HTR alone reads old hands reliably and LLM fluency masks errors.
Leo Team · July 20, 2026
Leo vs Tropy: Which Tool Fits Where in a Manuscript Research Workflow
Compares Tropy and Leo in manuscript research workflows, showing Tropy as an image-organizing metadata tool and Leo as the transcription engine that reads the text.
Leo Team · July 17, 2026
Why ChatGPT Struggles to Transcribe Handwriting from Historical Documents
Why ChatGPT struggles to transcribe historical handwriting, due to image downsampling and fluent confabulation, and when specialist HTR is safer than general AI.
Read the article
Decipher your archive
© 2026 Leo Technologies Limited. All rights reserved