I know there are folks here that hate voice assistants, I love mine, sort of. I love the function. I despise the fact that Jeff Bezos has little round spies all over my house.
I'd love to be able to along with my Hubitat rig, a voice assistant that is vendor independent, and honestly, anonymizes any online data so that I am not the product, or subject to inquiry because I looked up the ingredients for an unapproved soup recipe or something stupid like that.
I'd like as much as physically possible to run locally, but get useful data online like stream music, display weather etc...
I'd like all my switches, and controllers to just go in and work first time out of the chute.
I'd like more home automation that actually helps. I am giving a robot vac / mop some serious thought as well as a robot mower although my lawn guy would be quite displeased with me...
I'd like that voice assistant to actually understand me, and my wife, and... well extended family. We have a variety of accents including mine that is a midwestern / west coast, Texan thing and if I don't pay attention nobody understand me. I can literally go from "duuuude!' to "all y'all" without thinking about it and that messes with people, and Alexa...
The problem is you want the power of a multi-billion dollar massive cloud computing platform, but for $30, with no strings attached, all in a package that can fit on your nightstand. This technology simply doesn’t exist.
So how do they make it work? They give you a $30 device that taps into a remote multi-billion dollar cloud computing platform in exchange for massive amounts of data so you get it for “free.”
One could argue that the only way to have smart lights is to follow the same multi billion dollar massive cloud computing system, but we know thats not true. OP has a valid dream, it's just that the right people havnt put it out, yet. Its very doable and somebody somewhere will save us, someday.
False. Voice recognition is well established technology that can be run locally on very minimal hardware, I.E. a Raspberry Pi. The issue isn't it being possible, the issue is finding the right open source project and determining which hardware to get the job done.
The bigger issue isn't getting a voice assistant that is privacy focused at a reasonable price. The issue is integrating it with HE.
If I knew enought about coding, I would volunteer to do the needed tie ins between HE and at least 1 or 2 of the Open Source voice assistants. I haven't tested them yet, but from friends working with them for other projects, Mycroft seems to be toward the top of the heap, not to mention if you like Heinlein books the name is a real hoot too...
I don't hate voice assistants just am concerned about the ramifications of giving all my personal data away. I don't know what I don't know.. otherwise they are very handy.
The other thing about voice assistants is they dont have to be all that complex. Most people probably only have a few commands that they ever use anyways. Turn on lights, hows the weather? Whens my anniversary? My old 2013 truck can do basic commands like call so and so without needing internet access. A basic house assistant with basic commands should be doable and would work for 95% of what I would need.
“False.” I didn’t say voice recognition. He asked for a voice assistant. So running a neural net with vast machine learning capabilities. It’s fine if you want to disagree with me. But the. You follow up with “if I knew about coding”. I’ve been doing this for 20 years so I have a lot of experience and have even built some natural language processing tools. Yea things like Mycroft exist. Have you used it? It’s one thing to exist, it’s another thing to be good and useful (and meet the OP’s request of perfectly handling numerous accents and dialects). If Mycroft works for you, awesome. Personally I found it worse than nothing and quickly got frustrated with it.
There is also something like https://www.josh.ai/ however its not cheap either. I don't think anything that works well will be cheap.
There is a lot to voice assistants and a number of people have touched on it above. Understanding accents, different languages, logic, doing more than just simple commands (which the wife uses) such as asking for weather in various areas, unit conversion and other random questions.
There are also hardware considerations. These small boxes are more than just a voice command processor and a mic. They have processors and multiple mics to help them recognize who is talking, or help distinguish between a tv and a person or even where the voice is coming from to help with multi room installations.
On paper it sounds simple, but even Amazon and Google struggle to get it perfect.
This is why I gave up on it. They really struggle to understand my wife and daughter, but seem to work better with male voices. If my father comes over for some reason she jumps into the conversation constantly when he's speaking. In their current state of tech I just find them annoying honestly. They have a long way to go really. All I ever wanted was an easy way to switch lights on. All the other stuff they do is just fluff and kinda gets in the way some times.
I know with siri my friends got better results with changing it to UK english because of their accents.
With alexa, it took a while to get the room placement of my devices right. Then she started picking up things much better. My sonos beam is the best of them all though. I'm always amazed by it.
However, now amazon has changed something yet again, and I can say alexa in one room and all of them in the house will hear it.
Those random changes are the most frustrating part. Especially when you get it all set up right.
Yeah that was kind of the last straw for us. We had one in every room and the wrong one was constantly answering. Many times the one in front of us would ignore a question so the one on the other end of the house could respond. That used to work fine years ago, but it's been bad lately. In their defense my house is all hard wood floors and voices do carry, but again it once worked. The last few months we found ourselves just using cell phones to look stuff up and only using Alexa for light control. But then light control kept failing and requiring removing and re-adding skills to fix. It became a part time job.
But this all lead me to Hubitat so I'm grateful for that. A bunch of Echos are now sitting in a storage box and I'm rather enjoying the peace and quiet. Timers and motion sensors are just pure bliss now.
If you are into things Node-RED you can filter results by device and only have it respond if the command is issued from the device you want. The drawback is you need to create custom phrases like "Alexa, lights on" instead of "Alexa turn on lights".
I have a "good night" routine that only works if speaking to our Kitchen Alexa.
Some have it set so they can raise/lower shades depending upon the room. Very slick.
Sorry guys, the same sorts of arguments were used to argue that only mega companies could produce effective operating systems, yet Linux is out there and kiling it in the enterprise and embedded systems spaces.
I haven't tried an open source assistant yet, nor do I expect one to be as polished as say Alexa, although early variants of say Red Hat Linux were much rouhger around the edges than RHEL is nowadays... Time and effort...
And again, what's the harm in having a dream? If it actually came to fruition, we could all have that much more of our privacy back while having some good function.
And yes, voice assistants I have used seem to VERY MUCH dislike femminine voices, particularly ones with spanish accents.
Didn’t say there is harm in having a dream. Also didn’t say it’s impossible. I said with technology today, I do not believe it is going to happen. That said, technology jumps forward all the time. It will be possible at some point, but the neural net processing and deep learning systems needed to do this can’t be run on a small, local, machine... for now.
Edit: there is also a big difference between Linux and this. Linux is software dependent only. Everything Linux does was already “doable” on a home PC because Microsoft and Apple were already doing it. What you’re asking for here is hardware bound. No amount of open source wizardry will overcome the fact that the machines we currently have in our homes do not have the processing horsepower to accomplish this.
True enough. However don't write it off long term. Commodity hardware, even the exceptionally low end stuff is getting more powerful every single day. I could see a day in the not too distant future where a DIYer / open source project might be able to cobble together effectivley a private AI cluster out of some future iteration of Raspberry Pi type hardware, and the "Smart Speakers" simply access points...
I am sure there are plenty of folks that were saying home automation was too complex to run locally on a tiny pi sized hub not too long ago either.
Yup. Someday it’ll be possible. I’m sure you’re right. Just saying, for the time being, Alexa, Google home, and Siri are far ahead of what we’ll get out of an rpi.
Exactly! I've used voice recognition on PC's since the 1990's. The cloud was only necessary to make the smart speakers cheap, build a dependency on their infrastructure, and to monetize the input.
What you (and I) were using the in 90s were a few preprogrammed phrases with some very very dodgy voice recognition patterns. You had to get those phrases exactly right or they wouldn't understand you.
Even if one were to create their own machine learning cluster on the cheap, good luck getting the massive datasets necessary to train it. Also, I hope you have the time and energy to guide it along... It ain't just apt-get install and go.
Just to to YouTube and search for "Machine Learning Flappy Bird". Those are limited variables and a single objective with the computer generating it's own massive dataset (many many iterations). Now, translate that to language and the many things people want to do with their smart homes...
Dragon Naturally speaking was slow back then, but I would not have characterized it as dodgy. If by a few phrases you mean a modern dictionary, then ok. Teaching a machine to play a game is not the same as STT and TTS.
My old 2013 truck can do basic commands like call so and so without needing internet access. A basic house assistant with basic commands should be doable and would work for 95% of what I would need.
