Hacker News Reader

At https://snips.ai, we process all on the device, from the speech recognition to the NLU.

You can then use any language to create skills to make it perform actions.

Take a look at how to build a connected speaker with Snips: https://medium.com/snips-ai/building-a-voice-controlled-home...

There's some info here: https://mycroft.ai/documentation/mycroft-software-hardware/

Looks like it can all be done offline + OSS except for Speech-To-Text. But I think the key idea is even if it's using cloud for processing, you can at least trust it isn't listening when it shouldn't be (as long as always-on, wake word detection is offline).

Speech to text is precisely the most sensitive thing to protect. There is simply no way to offer real Privacy when processing voice in the cloud.

I’ve only played around with mycroft a bit, but I’m pretty sure the language processing occurs via cloud (home.mycroft.ai). As it’s open source, you can plug it into a local STT if you want to. The Snips.ai platform seems like it was designed for keeping everything local out-of-the-box, which is why I’m going to try it out this afternoon! After all, I really only want it to understand a handful of commands, mainly around audio and lights.