Recording redaction is now available for Voice API and TeXML. Sensitive details spoken on a call are silenced before the recording is stored, so card numbers, personal identifiers, and health information never reach storage.
What's new
- Three redaction kinds: pci silences card numbers, expiry dates, and security codes; pii silences names, phone numbers, addresses, and other identifying numbers; phi silences medications, doses, and conditions.
- Coverage across Voice API: enable redaction on recording start, dial, answer, bridge, transfer, and conference recording commands.
- TeXML support: enable redaction across call and conference recordings, including <Record>, <Start><Recording>, <Dial>. Attribute names vary by command.
- Redaction before storage: Access the redacted recording through Voice API webhooks or TeXML recording status callbacks.
- Current scope: works on English speech and is charged at the Deepgram speech-to-text rate.
Why it matters
A recording captures everything your customers say, including the details your compliance team is paid to worry about. Until now, the options were storage-side policies, third-party scrubbing tools, or accepting the risk: the sensitive audio still existed on disk before anything redacted it. Redaction happens upstream on our network, so sensitive information is redacted before the recording is stored, helping reduce the risk of exposing sensitive data. That closes the gap between "we recorded the call" and "we can share, analyze, or archive the call" without a data-exposure story in between.
Example use cases
- Payment and billing lines: capture card details over the phone without ever storing the number.
- Healthcare support lines: silence medications, doses, and conditions in recorded calls.
- Contact centers: support PCI and privacy compliance efforts by reducing sensitive information captured in stored recordings for quality and training.
Getting started
- Pick the kinds to redact: pci, pii, or phi.
- Pass redaction when starting a recording, or record_redaction when recording through dial, answer, bridge, or transfer commands.
- For TeXML, configure the relevant redaction attribute on your recording verb or command.
- Access the redacted file through the applicable webhook or recording status callback.
