Submitted via IRC for TheMightyBuzzard
Google has long been focused on artificial intelligence. Its Google Now and voice assistance projects have used AI to better the lives of users. The Google Home voice-based hardware unit brings its assistant to life, making traditional inputs and displays unnecessary. With just the power of your voice, you can interact with the device -- nothing else is needed.
The search giant has decided to take artificial intelligence to the maker community with a new initiative called AIY. This initiative (found here) will introduce open source AI projects to the public that makers can leverage in a simple way. Today, Google announces the first-ever AIY project. Called "Voice Kit," it is designed to work with a Raspberry Pi to create a voice-based virtual assistant. Please keep in mind that the Pi itself is not included, so you must bring your own. For this project, you can use a Pi 3 Model B, Pi 2, or Pi Zero. Want a Voice Kit? Here's how to get it. Heck, you might be getting one for free and you don't even know it.
Source: https://betanews.com/2017/05/04/google-open-source-raspberry-pi-diy-voice-kit/
(Score: 2) by kaszz on Monday May 08 2017, @07:10PM (10 children)
Seems it can run without any phone-home:
Voice Kit: instructions to build a Voice User Interface (VUI) that can use cloud services (like the new Google Assistant SDK or Cloud Speech API) or run completely on-device.
(Score: 3, Interesting) by takyon on Monday May 08 2017, @07:24PM (3 children)
Hmm, so perhaps terabytes of cloudy code and data aren't needed to make a useful virtual assistant with voice recognition? Control your own AI and let it talk to Google, APIs, etc.?
Reminds me of this:
https://soylentnews.org/article.pl?sid=17/05/04/1522245 [soylentnews.org]
[SIG] 10/28/2017: Soylent Upgrade v14 [soylentnews.org]
(Score: 2) by VLM on Monday May 08 2017, @08:54PM (1 child)
virtual assistant ... voice recognition ... AI
I've been playing with that kind of equipment as a hobby for many years and I'm not sure those three words appear together outside marketing.
My misterhouse and more recently openhab as a virtual assistant is very "victorian era servant seen and not heard" type of thing where there's a lot of if-then and cascaded and astronomically aware timers and alarms gone wild. But it doesn't "do" anything other than make my live better by doing a more disciplined job that I would do myself if I were not so lazy. I do have perhaps the most barque and bizarre super-smart thermostat ever seen.
My voice experience is Alexa actually does rock at telling me what time it is, or as a verbal jukebox, or timers and alarms I can set, but she's pretty useless otherwise. Integrating her with misterhouse involved a java app that pretended to be a hue lightbulb hub, and integrating her with my new openhab is apparently more complex. Eventually I'll be able to turn lights on and off by voice and stuff like that.
There's no AI involved in any of it.
I will say that comparing misterhouse and openhab, 20 years ago you got weird looking perl, and today you get total WTF DSLs for every little task just total WTF also python and java. The new stuff is not better than misterhouse, but the java interface supports more stuff, mostly kinda sorta.
(Score: 2) by takyon on Monday May 08 2017, @09:16PM
I have tried Alexa and I'm also not impressed. It's good at... playing radio stations (but the IP radio will cut out often compared to an FM receiver).
Kudos on finding Alexa's "backdoor entry".
I used the generic term "AI". Real life sentient/sapient machine brain slavery earns the upgraded term "strong AI". Regular AI could mean a lot of things. It could mean that the Alexa/Siri of 2022 does a lot better than the current version, but falls decades short of being JARVIS "strong". As it is cloud-based, you *may* not need to buy new equipment to enjoy "smarter" performance. There's a version history for it somewhere.
[SIG] 10/28/2017: Soylent Upgrade v14 [soylentnews.org]
(Score: 2) by butthurt on Tuesday May 09 2017, @12:27AM
review of this peach wreck ignition soft where four Linux:
https://www.lifewire.com/state-of-linux-voice-recognition-2204883 [lifewire.com]
(Score: 4, Insightful) by frojack on Monday May 08 2017, @07:55PM (3 children)
Seems it can run without any phone-home
Not usefully, is my bet.
Without access to a mountain of analysed voice samples I bet it is relegated to "bait" status when running stand alone. Maybe with a vocabulary similar to your dog, and capabilities equivalent to roll over and play dead dog tricks. The idea is to get you invested in the API enough to seek assimilation in the borg.
Google is really good at boot strapping one service to built another. Like Google Voice, a voice mail phone answering system (free of charge) that was used to get massive amounts of freely offered voice samples for voice reco development. That allowed them to walk around all of Nuance's [nuance.com] patents. I suspect some similar mining operation going on here.
No, you are mistaken. I've always had this sig.
(Score: 3, Interesting) by fyngyrz on Monday May 08 2017, @08:13PM (2 children)
Just as a data point, my 10+ year old, completely unconnected-to-anything GPS understands me pretty well. Understands Deb, my SO, a good bit better. Definitely imperfect, yet well into the "useful" realm of functionality. It also has a completely programmable wake-up (I've used "Yo, bitch" in the past, so that's solid) and it understands quite a few phrases and commands. It wasn't expensive, even back then.
Modern tech could do a lot better. Echo-level better, no, probably not, at least, not yet, but it's in the cards in the future (and proof of concept is us... we understand things pretty well if others aren't mumbling) and it's a lot better than having the device go stupid the moment it loses its Internet connection or the company providing the service goes away, changes its terms, or simply breaks the API so your device is now no more than a collection of raw materials.
There's a good potential candidate cooking up in the Mycroft project. [mycroft.ai] Open source; runs on minimal hardware, including the rPi. And the have an actual stand-alone device... although it's still in pre-order. Seems to be close to release, though, they're filling the kickstarter orders now if I understand the newsletter correctly.
(Score: 2) by kaszz on Monday May 08 2017, @09:03PM (1 child)
I'll guess the GPS voice recognition isn't available as open source.
What is "echo" level better?
The Mycroft project is very fuzzy about hardware demands. But their web page is all fluffy with bug pictures. But if it can run on RPi it should not be that bad.
(Score: 2) by fyngyrz on Tuesday May 09 2017, @01:04AM
Echo is Amazon's system. It is very good at recognizing speech. There's a command, "Simon says", where you can make it say anything. For instance, "Alexa, simon says, the problem we see here, my friends, is antidisestablishmentarianism" – it gets that and most everything else, right. As far as my experience goes, the Echo's by far the best at recognition right now. Could be stuff out there I don't know about, of course, but I've used Siri and Google's stuff, and it is very weak by comparison.
Yeah, if Mycroft will run on an rPi(2), it'll run on an even moderately current desktop or laptop with the right Python and other resources, etc.
I plan to get one. Amazon's "development" API for the Echo is an outright horror story, requiring canned phrases, utilizing zero actual AI past voice-to-text, either use of their "cloud" or instantiation of a dedicated secure server. After looking into it and realizing what a mess it was, I just settled back to wait. As sucky as that system is, someone had to beat it. Mycroft might be the one, as it's all open source so one would think that whatever its flaws might be, one could code around them.
(Score: 2) by edIII on Monday May 08 2017, @10:45PM (1 child)
That's very good. My first fucking question/rant here was going to be if Google is still using the fucking thing as a microphone to listen to goddamn everything.
I refuse to use any such technology precisely because it "phones home", which is really a misnomer. It's not phoning home, but acting like a thin client with Google's hardware doing all the heavy lifting. As such, Google gets to keep and use all the information obtained in such activities. Which is fucking horrifying.
I can't wait for AI, mastered speech to text, and neural interfaces, but only if they are free. By free, I mean Freedom included. That does mean running on local hardware freed from all binaries/blobs.
Technically, lunchtime is at any moment. It's just a wave function.
(Score: 0) by Anonymous Coward on Tuesday May 09 2017, @01:31AM
Okay.
You only ask questions or rant about what Google is listening to on the "fucking thing" because of your fucking?
Such kink! Google are fucking horrifying but who are you fucking? We cannot keep up.
Boner cancelled! I like my freedom and my local hardware running binaries - non-binaries and blobs not so much! Both tinfoil and latex hats will both be required in the future I see?
(Score: 0) by Anonymous Coward on Monday May 08 2017, @07:28PM (3 children)
It's the Soylent Family Assistant, similar to the Amazon Echo except designed to make family mealtimes, dinner parties and social gatherings into events to remember. Every time it can match a spoken keyword, it retrieves and recites comments posted to this site by Ethanol-Fuelled and TheMightyBuzzard.
May kickstarter it, any interest?
(Score: 2) by kaszz on Monday May 08 2017, @07:38PM
Everytime it recognizes a word or two it replays something it heard a while ago. "John is good" .. Raspberry-Pi response "Ooohhh John push it hard and spank me" ;-)
(Score: 3, Funny) by VLM on Monday May 08 2017, @08:45PM (1 child)
Can it stream Rick Astley "Never Gonna Give You Up"? I have an Alexa and I have been rickrolled thru an open window so if you're gonna be feature complete with the competition...
(Score: 0) by Anonymous Coward on Tuesday May 09 2017, @01:15AM
Definite possibility - added to the features list. I think the trigger for this should be anytime a non-soylent news channel is mentioned or vice-versa? Need TMB and EF input on the rickroll here...
(Score: 5, Insightful) by datapharmer on Monday May 08 2017, @08:19PM
Google should be ashamed of themselves for using this marketing tactic of "oh we are giving back to the maker community by letting them pay to access our closed source api"
Besides, this has already been done almost verbatim using open source by Mycroft:
https://github.com/MycroftAI/enclosure-picroft#readme [github.com]
They even offer a downloadable image for a raspberry pi 3 and a parts list...
Don't like it? Try Jasper instead: https://jasperproject.github.io/ [github.io]
Just want to do keyword recognition? Try PocketSphinx: http://cmusphinx.sourceforge.net/wiki/tutorial [sourceforge.net]
Google is just after free advertising again for putting their crap in cardboard boxes.
(Score: 3, Insightful) by jmorris on Monday May 08 2017, @08:37PM (5 children)
Why do they need all that crap? All they are doing is adding a speaker, microphone, light and button to a Pi and bundling you to their platform. It needs several boards for that? An amp chip plugged into the Pi with a microphone. The Pi has a headphone/mic port, why complicate things needlessly? We all know a standard headset can support a single button so they could drive everything into that one 1/8" jack except the light. One frigging GPIO pin can't be that hard for people as smart as Google, right? Talk about over designed!
(Score: 4, Funny) by LoRdTAW on Monday May 08 2017, @09:12PM (2 children)
This is the maker community we are talking about. The same people who think an entire OS stack with a few megs of JS code is how you turn a light bulb on and off.
(Score: 2) by kaszz on Monday May 08 2017, @09:30PM
Which is a result of their incompetency.
(Score: 0) by Anonymous Coward on Tuesday May 09 2017, @01:19AM
Me, I just use a switch. [scientificamerican.com]
(Score: 0) by Anonymous Coward on Tuesday May 09 2017, @11:01AM (1 child)
The Pi does not have a mic port. You have been misinformed.
(Score: 2) by jmorris on Tuesday May 09 2017, @09:27PM
Well crap. Saw it had a four pin 1/8" plug, should have read further... they only use it for composite video with no obvious capability for jack switching. One would think a SoC designed for a cell phone would have an mic/audio input, one would assume the brainiacs would have bothered to expose such a useful input, even if only on the 40pin header. One would apparently be wrong in ASSuming these things. :(