Hacker News Reader

Comment by Ajedi32 | Hacker News Reader

That's a ridiculous accusation. All that article is saying is that when you use the TV's voice control feature, your voice (and any other background noise) gets sent to their third party voice recognition provider (Nuance). That's just the nature of how voice assistants work, it has nothing to do with ads.

View on HN

Controlling a TV is not hard. I remember software which would run on a Mac Plus which would allow me to control the Mac using my voice. Are you telling me that a TV needs to connect to the internet to do that?

Depends on what sort of commands it supports, how well it needs to handle background noise, and how advanced their on-device voice recognition tech is.

For a company like Samsung, yes they very well may need internet access to do voice recognition at an acceptable level of quality. It's worth noting that both Amazon Alexa and Google Home also implement their voice assistant features that way. Google did recently move to on-device voice recognition on phones in the newest version of Android, but that was only possible to do without sacrificing quality thanks to some new advances in their machine learning tech.

The first Xbox One could control most hardware using the Kinect and simple commands like "Xbox On". The ability to understand language locally has been available since at least 1997 with the Dragon Dictate. The only real reason it's sent remotely is to learn from the data gathered.

Yes, we've had really crappy implementations of voice recognition for decades now. Doing voice recognition _well_ has been a somewhat more recent development (depending on who you ask it's still not good enough), and until very recently, most of the more accurate systems have either required extensive calibration to learn your voice, or been run exclusively server-side. The reason it's done remotely is because that's the easiest way to get accurate results; it's not some grand conspiracy.

I think the biggest improvement is the ability to understand the context of a sentence instead of just listening for keywords. I'm not saying it's a conspiracy because there's just so much data and lot of it probably isn't even useful. However, the recordings are definitely used to improve the ability to recognize the way different people say the same word for the purpose of improving recognition in the future.

ORLY? https://www.vice.com/en_us/article/8qbq5x/the-cia-spied-on-p...

That's a software vulnerability, not a feature. Also not related to advertising.

That's how they are designed, not how they work. There is little stopping manufacturers from doing voice analysis locally, they simply choose not to.

voice assistants do not require a network connection. disable wifi and cellular on your phone and “hey siri”, it works. if a company requires this info be sent to a server for processing there’s other reasons besides the device being unable to handle it

It's literally what Alexa was built for. Same with Google Assistant. They take the voice, use AI to process it.

Reply on news.ycombinator.com