Real conversations are messy. Redacting sensitive data shouldn't be.
Card numbers are interrupted, corrected and entered by keypad. SecurePass removes sensitive details from the original recording and returns redacted audio with a matching transcript. Test it on the calls your software actually handles.
Hear exactly what gets removed
A caller starts reading a card number, corrects a digit, then enters the remaining numbers using the keypad. Listen to the original and redacted versions of the same call. The sensitive information is removed while the rest of the conversation remains.
A billing call before and after SecurePass, with captions on the redacted spans. The video has sound.
Keep the speech workflow you already have
Already processing recorded calls? Send a sample recording to SecurePass through the API and inspect the redacted audio and transcript alongside your current process. You can evaluate the output before deciding how it fits into your application.
curl -H "authorization: $KEY" \
-F file=@call.wav \
https://api.autessa.com/clients/call-redaction/redact- Redacted audio, same length and channels as the original
- Redacted transcript with each span marked
| Type | Start | End |
|---|---|---|
| CARD_NUMBER | 0:07.7 | 0:21.2 |
| CARD_CVV | 0:29.2 | 0:31.5 |
| CARD ENTRY (payer channel) | 0:03.2 | 0:36.5 |
Supported audio
- WAV
- mu-law
- MP3
- M4A
- Opus
Single or dual channel, with keypad tones. Recordings sent to the API are deleted after processing unless you choose otherwise, and redacted results are kept for 30 days.
Test the difficult cases before you integrate
Straightforward examples are not enough. Try interrupted numbers, corrections and keypad input to see how SecurePass performs on the recordings that matter to your product. Review the output and processing options before you build it into production.
Interrupted numbers
The caller reads the card in four groups while the agent says "uh huh" between them. SecurePass treats the groups as one card number and silences each piece, while the agent's replies stay in the recording.
Sample call: one CARD_NUMBER span from 0:14.6 to 0:30.4 across the agent's acknowledgments.
Corrections
A caller says "sorry, that last one was" and repeats a group. The corrected digits are part of the same card number, so they are redacted too, not left behind as a stray fragment.
Synthetic test call: the caller restates "that last group was one one one one" at 0:17.8, and the corrected digits are silenced.
Keypad tones
On an automated line the caller never speaks the number. SecurePass finds the touch tones themselves and silences them, while the prompts and the pound key stay.
Sample call: account number, card number, expiry, code and ZIP entered as tones, all silenced.
Measured in English and Spanish
SecurePass is over 98% accurate on Autessa's own evaluation benchmarks across English and Spanish call recordings. On the PCI-specific benchmark, no PCI data leaked through redaction: not one full card number, expiration date or security code remained in the output.
The benchmarks are internal, so the best test is still your own audio. Send the calls your product actually handles and review the redacted output yourself.
- English
- Spanish
- 0 PCI leaks on the PCI benchmark
Autessa holds SOC 2 Type II, ISO 27001 and ISO 42001 certifications and supports HIPAA and GDPR compliance. Review its security documentation at trust.autessa.com.
- SOC 2 Type II
- ISO 27001
- ISO 42001
- HIPAA
- GDPR
- NIST 800-53 Moderate
- CJIS
- PCI DSS
- FedRAMP Moderate
Try it on a recording your software handles today.
Create an API key, send one call, and compare the redacted audio and transcript with your current output.