Redacting scans, spreadsheets, emails, audio and video

Why redacting scans, audio and video needs different methods

Personal data is stored differently in each format, so it has to be found and removed differently. A Word document holds text you can search; a scan holds only pixels; a recording holds sound.

LEXARIS supports 13 formats, including DOCX, PDF (with scans), Excel and CSV, JPG and TIFF images, EML and MSG emails with their attachments, audio and video. The same detection of more than 116 personal data types runs underneath; what changes is how the data is reached and how it is removed.

Scanned documents and images: OCR with burned-in redaction

A scan is a picture of a page, so the text must first be read with OCR (optical character recognition). Personal data is then detected in that text and the redaction is burned into the image.

"Burned in" means the pixels under the redaction are replaced, not covered. This matters because a common mistake is to draw a black rectangle over a PDF: the box hides the text on screen, but the original image or text layer often remains underneath and can be recovered by copying, or by removing the box.

Spreadsheets: rules per column

For Excel and CSV files, you set a rule for each column. Structured data is predictable, so choosing the treatment column by column is both faster and more accurate than treating the sheet as free text.

For example, in a customer list you might use:

• Name: token, so the AI can refer to customers and you can restore them later.
• Phone: mask, because the AI does not need it and you will not restore it.
• Customer ID: hash, so the same customer can be matched across files without revealing the ID.
• Notes: full detection, because free-text columns can contain anything.

Each of these modes is explained, with its reversibility, in seven ways to replace personal data and in the replacement modes section.

Emails with attachments

EML and MSG files are processed together with their attachments. Personal data in an email thread is often spread across the body, the signatures, quoted replies and attached files.

A forwarded thread may contain a dozen names and phone numbers in signatures alone, and the attached contract or payslip usually holds more. Processing the message and its attachments together means the personal data in attached files is not left behind when the email is sent to an AI. For the full AI workflow, see how to use AI with client documents.

Audio: transcription and tone replacement

Audio is redacted in two stages: the recording is transcribed with AI, and the speech segments that contain personal data are replaced with a tone or silence.

This suits call recordings, interviews, dictation and meeting audio. The transcript lets detection work on words, and because it is aligned with the recording, the matching stretch of audio can be replaced. The redacted file sounds like the original, except that names, numbers and addresses are bleeped or silent.

Video: tracking and blurring faces and plates

In video, faces and licence plates are detected, tracked from frame to frame and blurred. Tracking keeps the blur on a face or plate as it moves, turns or passes behind something.

Frame-by-frame detection alone tends to miss moments, for example when a face is side-on or partly hidden, and a single unblurred frame can be enough to identify someone. Tracking follows the object through the clip, so the blur stays in place between clear detections. As with scans, review the output before sharing, particularly in crowded or low-light footage.

Limits: which formats work where

Scans, audio and video are processed in Cloud only. With Keys stay with you, detection runs in your browser and handles text and documents that already have a text layer.

That is because OCR, transcription and video tracking need server-side processing. In Cloud, files are processed and encryption keys stored on our servers in St Petersburg, Russia (EU, UK, US and Swiss regions are planned and not yet available), and LEXARIS acts as a processor under a data processing agreement. With Keys stay with you, neither document content nor keys reach LEXARIS, in exchange for the narrower set of formats.

The supported formats and the Cloud and Keys stay with you comparison on the main page sum this up, and the guide to client-side keys explains when each option fits. Developers can automate the same processing through the REST API, Python SDK and CLI.