Skip to content

Data redaction operations

The Redact data job removes sensitive content from conversations that match redaction rules: matched audio intervals are replaced with silence and matched transcript words are replaced with a mask (by default *********). Audio redaction is destructive — the file is re-encoded (and re-encrypted where encryption is on) without the redacted content.

This page covers running and monitoring redaction. Rule authoring — pattern syntax, testing, and the review workflow — is covered by the dedicated Data Redaction User Guide, and the tenant administrator's view is in the Administration Guide.

Everything lives under Administration › Conversation Analytics › Data Redaction, with four tabs: Rules, Jobs, Redacted, and Highlighted.

The redaction job

The Jobs tab manages Redact data jobs. Besides the standard pipeline-job sections (data source, filtering criteria, post-processing, schedule), the redaction-specific settings are:

  • Action — one of three modes:
    • Redact data based on rules — match and redact in one pass (fully automatic).
    • Highlight moments based on rules — detect and record matches without modifying anything (non-destructive dry run / review mode).
    • Redact the previously highlighted moments — apply redaction to moments recorded by an earlier highlight run.
  • Rules — apply all Active rules, or an explicit list of Selected rules.
  • Limits — safety caps against runaway rules: Maximum redacted words per moment and Maximum redacted moments per call (both default 50).

The highlight-then-redact pair supports a human-review workflow: run a highlight job, review the detected moments on the Highlighted tab (entries there can be deleted to exclude false positives), then run a job with Redact the previously highlighted moments.

Because redaction alters audio and transcripts irreversibly, prefer this staged approach when introducing a new rule; a rule can also be tested against a single conversation without modifying files (see the Data Redaction User Guide).

History: Redacted and Highlighted

  • Redacted — the log of applied redactions: which conversations were modified and the moments removed.
  • Highlighted — moments detected by highlight runs that are pending review or redaction; rows can be deleted to prevent their redaction.

Pipeline placement

Redaction operates on transcripts (and audio), so the job runs after transcription — typically consuming a queue fed by the transcription job. When rules reference entity types, run entity recognition first. A completed redaction raises the Call - Redaction completed pipeline event, so downstream stages can be chained to run only on redacted conversations — see the pipeline overview.

Note

Rules are defined per tenant or at the System level (the Rules tab shows the tenant of each rule). The job honors its own scope and filters, so one global job can serve all tenants' active rules, or separate jobs can isolate tenants with special requirements.