Skip to main content

Protect sensitive content

Sensitive-content protection masks confidential values in transcripts before downstream keyword matching and analytics. It can protect recorded calls, live transcript output, and translations.

This capability and its protected categories are configured for your account by Speechlytics. Contact your representative to change the policy.

Protected categories​

Policies can cover common sensitive values such as:

  • payment-card details;
  • bank and financial identifiers;
  • government or personal identifiers;
  • contact information;
  • authentication credentials;
  • health-related identifiers;
  • other account-specific confidential values.

Exact detection depends on the account policy and the context available in the transcript.

How masking works​

Speechlytics uses two complementary passes:

  1. deterministic patterns mask well-formed values such as card-number candidates;
  2. contextual analysis identifies sensitive values that require surrounding language to interpret.

Masked content is replaced in the transcript rather than merely hidden in the interface. Downstream stages therefore operate on protected text.

transcription
    ↓
sensitive-content masking
    ↓
keyword matching, translation, summaries, analytics, and custom questions

This ordering prevents later analysis from receiving the original sensitive value. Depending on account policy, tags introduced by deterministic masking can also be removed before results are returned.

Live conversations​

For live calls, finalized segments can be masked before they are broadcast or stored. Interim recognition text should be treated as provisional because the final segment can be corrected and protected once enough context is available.

Result behavior​

When protection is enabled, call processing includes RedactionStarted and RedactionEnded lifecycle states. Final results contain the protected transcript. The original confidential value is not exposed through normal call-result endpoints.

If redaction cannot complete, subsequent processing may be delayed or omitted to avoid analyzing unprotected text.

Guidance​

  • Do not use transcript masking as the only control for raw recordings; apply appropriate storage and access policies to audio as well.
  • Test expected examples before relying on a category for compliance decisions.
  • Remember that transcription accuracy affects detection accuracy.
  • Avoid sending sensitive values in filenames or external metadata, which are outside transcript masking.