AI Audio Redaction Service
1. The Problem Statement
Currently, RingCentral Session Border Controllers (SBCs) capture raw audio and attach it to RingEX (REX) or RingCX (RCX). When these recordings move into RingSense/ACE, the system successfully redacts PII and PCI from the text transcripts, but the raw audio files remain untouched.
The Risk: Unredacted audio containing sensitive PII/PCI resides in REX/RCX for 30–90 days, and in ACE for up to 12 months.
The Liability: This exposes clients to severe compliance violations (PCI-DSS, HIPAA, GDPR), a risk compounded by the common client misconception that audio is redacted alongside transcripts.
2. The Proposed Solution
Develop and integrate a native AI Audio Redaction Service. This service will act as a mandatory processing gateway for recordings before they are committed to long-term or short-term storage.
End-User Control: Administrators will have the ability to select from standard compliance templates (e.g., PCI, HIPAA) or program specific PII/PCI fields to be targeted for redaction.
Mechanism: The AI will identify the selected spoken elements and overwrite those specific audio segments with white noise.
Outcome: A sanitized audio file that protects clients across all downstream platforms.
3. Proposed Architectural Workflow & SLA
To balance compliance with operational efficiency, the workflow introduces a brief, defined processing window:
SBCs / Recording Service: Audio is captured.
AI Audio Redaction Service (New): Audio is routed here first.
Processing SLA: The redaction service processes the file and applies white noise to targeted elements. The sanitized audio will be available within 3–5 minutes post-call.
Storage & Analytics: The sanitized file is then pushed to REX, RCX, or AIR Pro.
Long-Term Archive: The file moves to ACE for 12 months of compliant storage.
4. Legal Framework & Client Acceptance
Because this process fundamentally alters the original recording, legal and QA safeguards must be built into the onboarding flow:
Destructive Edit Acknowledgment: Clients must explicitly accept that the redaction is a destructive process—the original raw audio containing the sensitive data is overwritten with white noise and cannot be recovered.
AI Accuracy Disclaimer: The Terms of Service will require clients to acknowledge that while AI speech-to-text is highly accurate, it is not 100% perfect. Variables like poor audio quality, heavy accents, or crosstalk can impact detection, and RingCentral does not assume liability for occasional false negatives (missed redactions) or false positives (over-redacted segments).