I am working with an account that has voiced concerns that Deepgram's Speech-to-Text will pick up PII from conversations happening in the background. Does Deepgram have any documentation that I can share that explains how Nova-3's Speech-to-Text streaming avoids this? If not, are there any background noise cancellation/suppression features available?
#Noise Cancellation/Suppression
molten arrow
split spireBOT
Thanks for asking your question. Please be sure to reply with as much detail as possible so the community can assist you efficiently.
-# If you haven't done so, ensure your Discord and Github profiles are linked to Deepgram so you can earn points to redeem on cool stuff just by being active!
Hey there! You don't seem to have your Discord account connected to your Deepgram account. Not sure how you posted in here, but you should go to https://community.deepgram.com and link your accounts!
It looks like we're missing some important information to help debug your issue. Would you mind providing us with the following details in a reply?
- The programming language you are working in (e.g. JavaScript, Python).
- A request ID that triggered your error or issue.
tribal mantle
I am working with an account that has voiced concerns that Deepgram's Speech-to-...
We have PII redaction, but I would suggest your own noise cancellation prior to sending audio. It could affect transcript accuracy, but it is better than sending potential PII you're not aware of
molten arrow
Do you have documentation on deepgram PII redaction? Also any documentation on how noise cancellation affects transcripts and best practices on how to implement noise cancellation?
tribal mantle
The only source I have on it was a comment I myself made in our GitHub forum some time ago
https://github.com/orgs/deepgram/discussions/403
I can tell you our models are trained on the entire unmodified audio data. Noise cancellation changes the shape of that data. The models don't themselves "hear" anything.
To make models more accurate for NC audio, would need to train models on that data, which in turn would make natural language less accurately transcribed.
We believe the best quality transcriptions come from unmodified audio, with redaction applied, but that does not address the concern of sending PII.
Details about our redaction feature can be found here: https://developers.deepgram.com/docs/redaction