Transcription Endpoint

[Peter_Stoltenberg]

Hello,

I am looking to use v2.conversations.{id}.transcription for a custom Agent Assist component to get the real-time transcription and process it.

I had a question come up about PII/PHI. Is there any redaction of sensitive data as part of this real-time feed? Or, is it purely raw? In my tests so far, it seems to be purely raw, I just wanted to confirm. This also made me think about how the native Genesys Agent Assist handles that scenario. Anyone have any thoughts? @jacobshaw

Thanks,
Peter

[jacobshaw]

Hi Peter
Redacting sensitive information is a feature of voice transcription. It can be enabled as outlined here

Enable automatic redaction of sensitive information - Genesys Cloud Resource...

Users need the ViewSensitiveData permission to see transcripts that are not redacted. An overview of how voice transcription works with sensitive data here

Understand voice transcripts - Genesys Cloud Resource Center

[Peter_Stoltenberg]

Hi Jacob,

Just to clarify though: that only happens after the conversation is complete - for the transcript generated once the conversation is complete. It does not apply to the voice transcript topic (v2.conversations.{id}.transcription), correct?

[jacobshaw]

I tried this out myself and found that transcripts are coming through with PII not redacted. So from personal experience, the redacting happens later.

[jacobshaw]

After talking with the internal team, I got this info

Currently, for the near real-time transcription (emitted via kafka v2.conversations.{id}.transcription topic) the only way to redact PII is manually for the agent to do a secure pause, which mutes the audio while the caller utters their PII data, which is excluded from transcription. There is/was an effort to redact PII automatically via Anonymization/Named Entity Recognition (NER) service which looks like it's currently on hold.

[Peter_Stoltenberg]

Hi Jacob,

Thanks. That's what I noticed as well.

Appreciate you verifying!