Hacker Newsnew | past | comments | ask | show | jobs | submitlogin

Personalization would require training the network to recognize your hot word, which needs to work in all sound environments, and detect hot word spoken by different people. That network also needs to be optimized to be small enough to be run constantly by a low-powered CPU in standby mode.

I am not saying that it is not doable, it's just there are real engineering tradeoffs present here.



Yep, most people don't realize that:

1. Hotword detection runs on a minimal low powered local subsystem

2. The two provided hotwords run networks that have been trained on millions of samples in thousands of different environments.

That thing is very heavily optimized. Alexa and Bixbi get far more false positives and false negatives.


This. The Google home gets far less false positives and why the wake word makes far more sense. You also do not then get accidently recorded like with the Echo.


Never thought about that, but makes perfect sense. Thanks for the explanation!!


Aha. Interesting. That makes sense. Well, that's good to know. At least it means it's coming.

If it's primarily a technical challenge that raises the chances of this eventually becoming reality.


Personalisation would not give advertising of the company to any guests who witness you using the device ... that's my idea of the primary reason.




Consider applying for YC's Fall 2025 batch! Applications are open till Aug 4

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: