All products

AI transcription for scanned records

Make an image-only archive searchable while preserving page order, original spelling and uncertain readings. The scan stays alongside the transcript.

AI transcription for scanned records helps when a search misses a name you remember seeing on paper. With transcribe-source, your agent can read an image-only record page by page, add the text as a companion file and refresh your archive’s index. Finding a word then takes you back to the scan you need to inspect.

This is documentary transcription. A misspelling stays a misspelling. A doubtful reading carries a mark. If the new text conflicts with something already recorded, the agent can open a research question for you to settle. It leaves the existing claim in place.

What it does

  • Read an image-only source in its original page order.
  • Keep uncertain words separate from confident readings.
  • Extract an existing text layer when the source already contains one.
  • Find a phrase across records that previously held only pictures.
  • Record a discrepancy without silently changing the family record.

Install and use

  1. Install the skill and the Python-based Plaintext Family History toolkit. Its text-layer extraction uses pypdf. Image-only pages need an agent that can inspect images.
  2. Choose a private archive source. Ask, “Transcribe pages one and two, keeping uncertain readings visible.”
  3. Search for a distinctive word, then compare its mark and spelling with the original page.

FAQ

Source

Made by asielen. Original skill: transcribe-source, published on GitHub under the MIT licence.

View on GitHub