News for hackers. Stuff that matters.
Re:Apple suspends Siri response grading in response to priva... by londons_explore on Friday August 02, 2019 @04:16AM (#20591385)
> We really need mandatory ethics courses in CS.

Voice recognition isn't really possible without someone listening to train the system.

If the option is "don't have a voice assistant at all" or "have a voice assistant that sometimes has a human listening", I suspect the public would mostly choose the latter.

When you call a company helpline and get connected to a human rather than a robo-voice, most people are happy about it. It's curious they treat voice assistants the other way round.

Re:Apple suspends Siri response grading in response to priva... by icebraining on Friday August 02, 2019 @05:27AM (#20591650)
> If the option is "don't have a voice assistant at all" or "have a voice assistant that sometimes has a human listening"

It's false that those are the only choices. You don't have to use the whole pool of users as unwitting guinea pigs of your system. You can use a voice assistant that was previously trained on a well informed and hopefully paid pool of volunteers.

Re:Apple suspends Siri response grading in response to priva... by cameronbrown on Friday August 02, 2019 @11:01PM (#20598381)
The trade-off there is obviously due to logistics your pool of volunteers isn't going to be as representative of a global population so training data quality is going to be worse.
Re:Apple suspends Siri response grading in response to priva... by oulipo on Friday August 02, 2019 @05:09AM (#20591586)
This is what we do at https://snips.ai, we are building the first 100% on-device and private-by-design Voice AI so that you don't have to choose between having an assistant and having your privacy!

It works in english, french, german, japanese, spanish, italian, and more coming soon! It is 100% free for makers!

Re:Apple suspends Siri response grading in response to priva... by ctz on Friday August 02, 2019 @04:31AM (#20591444)
> Voice recognition isn't really possible without someone listening to train the system.

IBM ViaVoice was released in 1997, and did standalone voice recognition to an acceptable degree from my recollection (after a couple of hours training).