From a file to a book that reads itself aloud.

Seven steps. You do the first and the last.

The seven steps

  1. 01

    Upload

    Drop in a file. Several can wait in a queue.

  2. 02

    Classify

    Digital, scanned or tagged: each goes down the path that suits it.

  3. 03

    Rebuild

    Text, headings, lists, tables, figures and equations become one structured document, with a reading order.

  4. 04

    Describe

    Figures get short and long descriptions.

  5. 05

    Check

    25 checks, and confidence scores for every part.

  6. 06

    Fix

    One-click fixes, edits in place, or “leave it as it is”, with a reason recorded.

  7. 07

    Listen and publish

    Run the Listen Check, then download HTML, EPUB 3 and XML, or open the book in the reader.

Confidence scores for the scanned sample: overall 98%, OCR not applicable, layout 100%, table 100%, figure 90%, structure 100%.
The report card for our scanned two-page sample. In our run it took about 30 seconds from upload to outputs.

Rules that keep a document honest.

Missing text is counted, never hidden.

Text in the source that isn’t in the result is reported. On our own sample it found two lost lines and listed them first.

A weaker reader is never used silently.

If the chosen reading method fails and another runs, the document says so.

Doubt is shown, not guessed.

Text it can’t vouch for is flagged for review.

Your decisions stay.

Listen Check decisions are re-applied every time a document is processed again.

What you need

Hosted

A browser.

Nothing to install.

On-premise

A Linux server with an NVIDIA GPU, and Docker.

We set it up with you.

Bring a document to the demo.