Format · Call transcripts
Contact-centre transcripts feed quality reviews, analytics and voice bots. They are also where card numbers, security answers, names and addresses end up as plain text, read out by customers in whatever order they like.
Treat them as free text, not fields. Detect names, card numbers, expiry dates, security answers, addresses and account numbers, including numbers read out in words, in pieces or corrected mid-sentence. Remove card data outright: PCI DSS v4.0.1 requirement 6.5.5 keeps live card numbers out of pre-production unless that environment is part of the cardholder data environment and protected as such. Replace names and addresses with consistent synthetic values, so each speaker stays one person across the conversation, and keep speaker turns and timestamps so the transcript still works as test input.
Then check the output against a labelled sample, because speech-to-text errors can defeat pattern matching, and handle the audio recording separately: masking the transcript does nothing to it. For chats and tickets, see chat log masking.
Card security codes may not be stored after authorisation, even encrypted, and the PCI Security Standards Council confirms that includes call recordings. PCI SSC FAQ 1210
No. Synthesized works on text. Audio redaction belongs with your speech or contact-centre provider.
No single tool makes you compliant. Masking transcripts removes card data from one place it collects; recordings, logs and your assessor's view of the whole flow still matter.
Plain text and JSON transcripts with speaker turns, plus the platform exports you use, scoped on the kickoff call.
Documents and text use the same policies, the same engine and the same deployment as these results.
Published results from database and synthetic data work on the Synthesized platform.
Bring 50 transcripts from one export. In 48 hours you'll see them masked, with the evidence.