[starsinspace]

It's interesting how they talk about "doing as much on device as possible"... but Siri's voice recognition still works by sending the Siri request audio to an Apple server and doing the voice recognition there (only the trigger phrase "hey siri" is recognized on-device). Why isn't it all done on-device? I'm pretty sure even older iPhones have more than enough CPU-power for that.

[GeekyBear]

Isn't this something that only recently has become possible, and still has an accuracy cost?

[starsinspace]

iOS already has a separate feature "offline dictation" which works on-device. I don't understand why that isn't used for Siri.

Also about "recently"... back in the late 90s I had a desktop PC running Windows 98, I think it was a Pentium 166 MHz with 32 MB RAM. On it, I had a voice recognition program called "Dragon Naturally Speaking". It required a little training with my voice but after that, it worked remarkably well. And that was over 20 years ago on a PC with a - by today's standards - very primitive CPU. Decent voice recognition isn't new technology.

[GeekyBear]

The problem with Dragon was that it was so inaccurate that a moderately skilled typist could produce final corrected text much more quickly than they could dictate text and then make corrections.

Looking online, this is a feature Apple is adding in this year's iOS and Google is adding to Pixel devices only so far, so I would expect to have to give them some time to get on device speech recognition working at all as a first step.

[jbverschoor]

Exactly. Also, smart Siri is so stupid, it recently started correcting the contact I want to call to someone else, even though it was properly recognized and transcribed.

Funny thing is I call this person daily, and the other one was 6-7 years ago

Frustrating as hell