Documents and OCR

A document dataset turns PDFs, scans, office and text files, and audio recordings into Markdown text that you can read, correct, review, export, and use in Annota Brain knowledge bases. Each document is split into pages, so you can check the extracted text next to the original.

Supported formats

Choose Documents as the dataset type when you create a dataset. Document datasets accept these files:

Kind	Formats	How Annota processes it	Credits
PDFs and scans	PDF, PNG, JPEG	OCR reads each page as an image	Per processed page
Office and text	TXT, Markdown, DOCX, ODT, RTF, HTML, EPUB, PPTX, XLSX, CSV, XML	Converted to Markdown, no OCR	None
Audio recordings	MP3, M4A, WAV	Transcribed in parts	Per started minute of audio

Limits:

Each file can be up to 100 MB.
A PDF can have up to 1,000 pages.
A converted office or text file can hold up to 1 MB of text.
Legacy DOC, PPT, and XLS files are not accepted. Save them as DOCX, PPTX, or XLSX first.

See Upload images and documents for how to add files.

How processing works

Annota queues every document for processing as soon as it is uploaded. You can leave the page while it runs.

PDFs and scans: OCR reads every page as an image, also when the PDF has selectable text. The result is Markdown with headings, lists, tables, and math. Images and charts on a page are saved as figures with a short machine-written description, which is not part of the document text. When all pages are done, tables that continue across pages are joined on the first page.
Office and text files: Annota converts the file to Markdown and keeps only the converted text. Spreadsheets and CSV files become tables, and XML becomes a code block. Long text is split into short pages.
Audio recordings: the recording is transcribed in parts of up to twelve minutes. The first page, Overview, summarizes the whole recording. Timestamps in the transcript play the recording from that moment.

The Processing column of the asset list, and a badge on each tile in the grid, show where each document is:

Status	Meaning
Queued	The document is waiting to be processed.
Processing	Pages are being read. Progress updates as each page completes.
Ready	Every page was processed.
Partial	Some pages were processed and others need another attempt.
Failed	No page could be processed.
Paused	Processing stopped because the organization ran out of credits.

If processing takes at least a minute, the person who uploaded the document gets an email with the outcome.

Review a document

Open the dataset, select Assets, and click a document.
Compare Original document on the left with Extracted text on the right. Use the zoom controls or Whole page to fit the original.
Move between pages with Previous page and Next page, and between documents with Previous document and Next document.
When the whole document is checked, set its review status: Approved, Rejected, Pending review, or To annotate. The status applies to the whole document.

Office and text files show the extracted text only, since their original is the converted Markdown. Select Download document to download the source file.

Correct the extracted text

In Extracted text, switch from Preview to Edit to see the Markdown.
Correct the text. Press Ctrl+F (Cmd+F on Mac) inside the editor to find and replace.
Select Save page, or press Ctrl+S (Cmd+S on Mac).

Moving to another page or document, or leaving the review, also saves your edits. Check the Unsaved changes and Saving changes… indicators before you close the tab. Edits change the extracted text, never the original file. Corrected pages are never processed or overwritten again.

For all review shortcuts, see Keyboard shortcuts.

Retry unfinished pages

When a document is Partial, Failed, or Paused, the review screen shows which pages need another attempt. Completed pages stay available and are not charged again.

Save any edits first. Retrying is disabled while you have unsaved changes.
To retry one page, choose it in Page to retry and select Retry page.
To retry every unfinished page, select Retry unfinished pages. For a recording, select Retry transcription.

A paused document needs credits before it can continue. See Manage your plan.

Credits and storage

OCR uses credits per processed page, and audio transcription per started minute of each part. The review screen shows the cost before you retry.
Converting office and text files is free, and so is retrying them.
Processing starts only while your organization has a positive credit balance. A page is charged once OCR has read it, even if a later step fails. Pages that fail before they are read are free.
Original files, extracted text, and figures count toward your organization’s storage.

Current rates are on the pricing page. See Credit usage for how usage is reported.

Next steps

Dataset versions: freeze the reviewed text.
Export datasets: download Markdown or an LLM Wiki.
Knowledge bases: ask questions about your documents in Annota Brain.