tgauchat
(ActionTiles.com co-founder Terry @ActionTiles; GitHub: @cosmicpuppy)
21
No offense, but ironic that your post starts by saying the SmartThings Hub should do more local execution, while concluding with examples of reliable cloud services.
But that’s the point: Samsung envisions that the SmartThings Cloud (including the internet infrastructure to households) will, soon enough in their timescale, be as reliable as all other modern utilities (electricity, water, gas, cable TV, and … internet).
Plenty of businesses/products absolutely depend on “the Cloud” as the critical piece of their business: Gmail, Facebook, Netflix, Hulu, YouTube, … Amazon, eBay…, absolutely every cell phone service provider, …, …
So why not Samsung? They don’t care if it isn’t “utility quality” today; they care that they will be there in 2, 5, or 10 years.
The original SmartThings was conceived by people who weren’t engineers. (Indeed, for years the CEO continued to use an example of what he wanted to accomplish something that could never be achieved with SmartThings as it was sold – – notification of a power outage.)
As I said at the time, although I don’t have any hard information for this, I believe that Samsung purchased SmartThings for the name and the positive buzz it had back in early 2014, because it kept them in the press as an IOT leader while they were working on their own projects, which were then several years away from being consumer-ready. They did have a cloud in mind, but it was the Artik cloud. I believe that’s the trajectory they’ve been on, and the one they will probably achieve in a year or two.
The SmartThings staff is and has been aware of the reliability and stability problems for quite some time and they share a lot of the same sentiments as customers and community members. We may have finally hit that point that begins to turn things for the better once again.
For those interested, here is a post from @vlad (ST Staff Member) a few hours ago:
I’m ok with a reliable cloud for some non essential things, like music from my Amazon Echo. But, I choose not to store important data files of mine in the cloud, and I use very few web based apps - other than the obvious browsers :). But when it comes to home automation (think thermostats and security) I want damn reliable performance. Not in 5 to 10 years🙀. To sell a product/service that’s so immature- That’s deceptive marketing if you ask me…
Hmmm - I wonder if Samsung realizes that I, probably like many others, will only dabble with ST considering its current state and known unreliability? There’s no way I’m changing my 5 WiFi thermostats (2 properties) to being managed by ST at this time, or my alarm systems or …! At this point it’s got to just be a hobby - which means I’m not buying any other HA goodies they might offer… So which is worth more to Samsung - data mining or sale of physical items?
After doing over 30 yrs of programming, my take is that all software should be as reliable as possible. The more external services (cloud) that a product like ST relies upon, the more holes in its armor! And think about it: for ST to function, the local layer HAS to be functioning, so USE it for 100% of the HA activity. Use the cloud as a backup and sync resource… Just sayin…
Totally agree too, but I think they’ll probably focus on making their cloud less of a mess, which will also improve reliability. And hopefully they’ll continue to make modest progress with local processing too. I just don’t think we’ll ever see it become a core priority for them.
Contrast that with the guys who started the company that’s making the hubitat hub. Local processing is one of their primary reasons for creating that hub.
“Contrast that with the guys who started the company that’s making the hubitat hub. Local processing is one of their primary reasons for creating that hub.”
Hmmm, I wonder why they decided local operation was a priority . Experience is the best teacher…
Speaking of a good use for cloud processing: Amazon has nailed voice recognition with all their speech aware device Software! Same with Apples Siri.
Now - contrast that with my brand new Samsung The Frame TV - it’s voice recognition it abysmal . I’m pretty sure the voice processing is being done in the Samsung One Connect unit- totally local from what I have observed. Here’s a case where they should defer to cloud processing, but for some reason haven’t? Maybe they know their cloud abilities and reliability aren’t up to the task? In this case the cloud should be the primary voice processor, and local processing secondary.
Actually, if I was designing/programming it, I would process voice locally first, but offer an “oops” button or voice command to let system know it did not understand my voice command. It would then send the captured voice data up to the cloud for more robust analysis - assuming their cloud truly was more powerful and accurate🤔
True, but those guys seem to have a more specific market in mind of hobbyists/enthusiasts. While it’s far from a finished product, they don’t have a mobile app or any built-in method for remote access. That’s because their intended audience is comfortable logging into the hub from a browser window, and can probably VPN into their LAN if they need to access something while away. But everything that can possibly run locally does in fact work that way.
The more general consumer market has different priorities, and Samsung might be focused more on that (ie make the cloud work better so reliability is acceptable, rather than pivoting to prioritize local processing).
As others have pointed out, there are cloud-based systems that work more reliably than ST.
The obvious issue with this order is processing time. Ideally, both local and cloud processing should happen simultaneously, so the local system can “check its work” against the cloud response, so when the cloud is unavailable, the local system would be as close to seamless as possible.
I follow voice recognition technologies very closely as I am quadriparetic and dependent on them.
I don’t know of anyone doing local voice recognition except for dragon which requires a laptop level of processing power and for the person to be using a microphone within about 6 inches of their face.
Everybody else, including Samsung’s smart televisions, is using cloud for voice recognition. The processing requirements are just way too high if you want natural language recognition.
For example, from the manual for the frame TV:
The {voice interaction} function is available only when the TV is connected to the Internet.
You can still do voice recognition badly even if you have a cloud. And there still a huge local hardware component in the microphones being used, which is where the echo first dominated.
But almost all voice recognition for more than 4 or 5 specific words other than the laptop option mentioned is now done in the cloud.
Agreed - if one can provide excellent local voice processing - do it. I’m guessing it still takes a lot of processing power to get voice right (accurate and fast). I must say though, Amazon definitely has it figured out! I’m just puzzled why all voice systems don’t allow you to give feedback and “teach” them to better recognize MY voice. The closest is nav systems in cars, where they’ll display a number of items (e.g. addresses or songs) that the software thinks you said.,. Isn’t that what good AI is all about🤔
Interesting info - thanks! The Samsung Frame’s recognition was just so poor I assumed they were using underpowered local hardware for processing. The fact that they are using their cloud doesn’t bode well for the future of ST
JD - how are car manufacturers able to provide reasonably good voice recognition using only local hardware? Of course your cars voice mpg may be different than mine (Honda Ridgeline). But my truck does a respectable job accurately and reasonably quickly processing voice commands…
. Experience is the best teacher…
. I’m pretty sure the voice processing is being done in the Samsung One Connect unit- totally local from what I have observed. Here’s a case where they should defer to cloud processing, but for some reason haven’t? Maybe they know their cloud abilities and reliability aren’t up to the task? In this case the cloud should be the primary voice processor, and local processing secondary.
