Conversation Redact: chats, captions, and audio

Conversation Redact is included with Cases plans. It is on with the plan. Outbound Guard seats do not include it.

What you can upload

• Microsoft Teams JSON, Slack export zip or JSON, WhatsApp chat text.
• WebVTT, SubRip, or a generic speaker transcript.
• Audio: WAV, MP3, M4A, OGG, FLAC, WebM. A zip may contain audio plus a caption file.
• Video is rejected.

How review works

Northdocs parses the file into speaker turns, runs the same deterministic personal-data detectors as document review (including name gazetteer, narrative cues, and speaker display names), and opens a conversation review page instead of the PDF canvas. Click a timestamp to seek the recording. Accept a person to wipe that name in turn text and in speaker labels.

Accept or skip each identity once. Select a missed name in the transcript and confirm Redact to redact it everywhere it appears. The prompt shows how many mentions will be marked, and Cancel or Escape leaves the transcript unchanged. A confirmed name is accepted right away, along with every mention written the same way or capitalized. A mention with other capitalization, such as the Danish word han next to the surname Han, is added for you to accept or skip. You can also mark the speaker label. Click a highlighted name to hear that spot in the recording, before and after Finalize. For recordings, choose beep, silence, or voice anonymize on accepted time ranges. Forensic light, standard, and strong shift vocal-tract length, pitch, and bandwidth and add masking noise. You can also anonymize a speaker across the whole file.

Each spoken name is timed from word-level speech-to-text, and the treated range starts a little before the name and ends a little after it. When a caption file comes with the recording and AI / ML detection is enabled, Northdocs aligns the captions to word-level speech-to-text the same way. Without word timing (AI off), the range is widened across the caption cue so the whole name is covered, and the review page says which turns use estimated timing.

Allowance

Chats and transcripts count toward the workspace page allowance (one page per 15 turns). Audio duration uses a separate voice-minute allowance per user: 60 during Trial and on Starter, 180 on Pro, 600 on Firm, unlimited on Enterprise. A captioned recording can consume both. The workspace voice pool is the per-user allowance times billed seats, with the same 5% soft cap as pages.