Hacker News new | ask | show | jobs
by relaunched 2271 days ago
I'm interested in knowing more about why closed captions would imply not end-to-end encrypted. Wouldn't it be possible to build a model and distribute the model with the client-side application, and run it at the edge?
1 comments

If they did that, everyone would have the model (meaning you would see closed captions in a lot more places, because it would absolutely be stolen).
Google translate and Google Gboard offer offline voice to text transcription... So it would seem the model is indeed on your device, and Google says as such - http://ai.googleblog.com/2019/03/an-all-neural-on-device-spe...
On the ARM TrustZone, maybe? Or whatever co-processor run Widevine.