Hacker News Reader

Comment by phkahler | Hacker News Reader

The notion that voice recognition is still hard enough to require doing it on a server somewhere still bothers me a lot. IMHO there should be no excuse for this today. I understand that the device needs to be connected to the internet anyway, but this whole thing is such a gimmick it's not worth the privacy risk.

[edit] to clarify I think speech to text can be done on a device today, that's why this bothers me.

View on HN

> voice recognition is still hard enough to require doing it on a server

I'm not sure this is true, it's just that since they are phoning home anyways, why commit to extra local hardware to change sound bytes into text words?

I would love to see an offline version of one of these with an easy API (do any high quality ones exist?). If I then wanted a generic Google request to give me the first Google response, so be it.

The awesome folk at Snips are challenging the idea that you can't do it locally https://snips.ai/technology/

Nice. I wish every route I took there didn't require me to sign up. But I did find https://github.com/snipsco and am browsing now.

> is still hard enough to require doing it on a server

Hot word detection would suggest that voice is easy. "Accent" and "Intent" recognition is the hard part.

Not sure what you mean exactly but this seems wrong. These devices do hotword (and speaker) recognition largely locally, but speech to text is done server side.

Reply on news.ycombinator.com