Home Automation Domotica Forum Europe, Bwired Forum

Digit wrote:I advise you to read a little bit more about the Homey project before judging too quickly, cause some of your statements are incorrect.
For instance, Homey will be based on the Raspberry Pi Compute Module, not on a Raspberry Pi Model B as you might think.
It's even mentioned on their Kickstarter 'Home page', hard to miss I'd say. Compute Module means no SD, but eMMC.

I did read over that indeed Very good point!

Digit wrote:And what's the problem with receiving the same signals in multiple places? I don't see the problem.
I'm sure they'll fix this 'problem' with EUR 170000 in their pocket.

Well that 170k is primarily for getting the device developed further and making it more stable; but indeed as they do have it in their plans it is likely they can resolve this, it can be a tricky thing though as sometimes you might want the redundancy. Eg does a signal received at two receivers mean it was sent twice (thus repeated, eg twice pressing the same button); or just a duplicate of the same transmission?

Digit wrote:Personally, I'm more interested in a "private / local" speech recognition solution

But for that they are using the Google Voice recognition stuff; I don't think they will develop anything of their own there. Doing their own voice solution would require them to hire quite a few more people who actually understand that stuff, and there are not many folks out there that do.
The Jasper system already demonstrates that, and that where folks who spent a lot of time on it.

>, combined with the best TTS there is - no matter whether it falls under PA or not.
> Who cares about what I let my computer say to me (the News, weather forecast, yes master, 20 unread emails, ... ) ?[/quote]

For me it is not the fact that it is something that somebody will read along (that was just a side-note as they make a claim about it); it is more to the fact that I don't let any of my sensors talk to the Internet; they have no business doing so and the VLAN they live in does not have routing towards/from the Internet.

More importantly the depending on an external service is bad: Internet down, and you can't turn on your lights the comfy way you are used to.

Digit wrote:The reason I backed Homey is (for me, and I think for the majority) that voice is the ultimate way of interacting with a HA system.

I cannot disagree, that is absolutely awesome

Digit wrote:"Jasper" is funny but not good enough for me - TTS is very bad. But Jasper was the trigger for me to revisit Voice Control once more:
http://blog.hekkers.net/2014/04/16/home ... e-control/

Yep, I've read that, very good writeup, but there you also resorted to using the Google API in the end as the self-contained system is just not good enough (and unfortunately I don't know any alternatives to it either...)

Digit wrote:And I've seen some "early work" on Homey (before it had that name) on Tweakers.net and was flabbergasted with how well it performed.
So when Homey appeared on Kickstarter the only thing I could do was to support their effort to make Homey work.

Nothing more, nothing less. I just hope they succeed in creating an awesome product, open enough to embed in our own systems.

I understand that sentiment

I wish them good luck too; it does not seem to be the tool for my job though.


Btw, as you already have a RFXLAN and now that new RFXtrx thingy, what part of the 433mhz will you let handle by the Homey, or will you make the RFXLANs sent their output to the Homey so that it can do all the centralized things it needs?

The advantage of having multiple 433mhz receivers is ofcourse that one can spread them throughout a house and avoid distance/wall issues. for reception.

But for that they are using the Google Voice recognition stuff; I don't think they will develop anything of their own there. Doing their own voice solution would require them to hire quite a few more people who actually understand that stuff, and there are not many folks out there that do.
The Jasper system already demonstrates that, and that where folks who spent a lot of time on it.

Yep, they will probably take the Google Speech API road. Not much info to go on, but it 'sounds' like it.
I don't know if that is a wise choice actually
Google Speech API v1 has been upgraded to v2 but it has some limitations (50 requests per day and no way to change that) and it seems that a lot of people are getting 403 replies lately (I'm one of 'em).
That made me wonder how long this API will be available; cause from what I've read it seems to be just a developer tool, in fact not needed anymore now that there's a HTML5 Speech API around.
Chrome already has its Web Speech API since version 25. For now, I think the best bet on using Google to do Speech recognition on a RPi-ish machine is waiting for node-webkit to support it?
Or maybe do some reverse engineering on how Chrome does 'it'... might be a tough job too. So right now, I'm fiddling with pocketsphinx to see if I create something useful.
I'm confronted with language models, dictionarys, bla bla... a lot of things to figure out, hopefully worth the time spent...

More importantly the depending on an external service is bad: Internet down, and you can't turn on your lights the comfy way you are used to.

I agree... life becomes very hard without Internet these days - a lot of things won't work anymore. Phone, mail, social stuff, banking, ...
But loosing speech still keeps the less-comfy ways of doing things up & running. It ain't that catastrophic to call it a wrong to do - unless you don't have alternative ways, of course.
Internet down is not my main concern - relying on temporary services is.

Yep, I've read that, very good writeup, but there you also resorted to using the Google API in the end as the self-contained system is just not good enough (and unfortunately I don't know any alternatives to it either...)

You're right, I did, it was too easy not to explore this But not for long. I put that project on hold for a while, searching for a good mic.
And once I had one (about 2 weeks ago) and played with it, Google Speech API gave me the 403's. I gave up on it.
Since a day or 2 I'm reading/learning about pocketsphinx and how to make it behave like a dutch

Btw, as you already have a RFXLAN and now that new RFXtrx thingy, what part of the 433mhz will you let handle by the Homey, or will you make the RFXLANs sent their output to the Homey so that it can do all the centralized things it needs?
The advantage of having multiple 433mhz receivers is ofcourse that one can spread them throughout a house and avoid distance/wall issues. for reception.

Homey will handle none of that. My own system is and will stay the 'core'. (where there is no 'core' actually, it's just a bunch (>30 or so) of processes producing and/or listening to MQTT messages).
I hope I'll be able to use Homey as an additional interface that supports speech, but I don't want it to also read sensors or control lights or whatever.
I just want it to send me (ehh, my HA system) what words were recognized - my system will take care of the rest.


PS
Regarding multiple RF receivers. I have a single driver that connects to multiple RFXCOM receivers (2, soon 3) and this driver collects all the information those receivers have to offer.
All that info is filtered on duplicates and only changes will result in messages to my MQTT broker.
So in fact it doesn't matter whether I have 1 or 20 receivers - the load on the output side of the driver (= MQTT messages) stays almost the same.

Digit wrote:Yep, they will probably take the Google Speech API road. Not much info to go on, but it 'sounds' like it.

heh, "sounds like it"

Digit wrote:I don't know if that is a wise choice actually
Google Speech API v1 has been upgraded to v2 but it has some limitations (50 requests per day and no way to change that) and it seems that a lot of people are getting 403 replies lately (I'm one of 'em).
That made me wonder how long this API will be available; cause from what I've read it seems to be just a developer tool, in fact not needed anymore now that there's a HTML5 Speech API around.

And that is the interresting thing about Homey; their primary coolness comes from two things: voice control + lots-of-radios, if the first thing (voice) goes away, there is not much coolness left as the lots-of-radios most people who build things themselves already have, and also, it then becomes a bit of a brick unless one is going to use good old buttons again.

For them to sell Homey with "voice control" requires for them to come up with something that can last.

They claim they only rely on "their" servers, which are "in the Netherlands", only for being able to connect to the device when not in the local network, thus as a rendezvous service; they could solve that easier with a number of other methods though. (but maybe they are relying on simple things like TURN or IPv6... no detail on how that is done is known)

But they apparently, for the voice, rely on a service that is not only indexing everything you do, is controlled by a US entity and which is a service that can just go away. I think that is a bad move. I thus can only hope they have a backup plan for that.

Chrome already has its Web Speech API since version 25. For now, I think the best bet on using Google to do Speech recognition on a RPi-ish machine is waiting for node-webkit to support it?
Or maybe do some reverse engineering on how Chrome does 'it'... might be a tough job too. So right now, I'm fiddling with pocketsphinx to see if I create something useful.
I'm confronted with language models, dictionarys, bla bla... a lot of things to figure out, hopefully worth the time spent...

RPi has a very tiny CPU. Yes, it can do 1080p video, but that is merely due to the hardware acceleration, there is no hardware acceleration for voice recognition...

The Web Speech API, is exactly that, just an API to the Google service. Hence, that has the same problem: external service that can go away (not even considering the privacy implications or response time), or as my lady recently reminded me: when the shutters are down and the power is out we can't open anything... except the front door and elevator, which then is not working either, but when there is fire there, you can't even get out of the house through the terraces as the shutters are nice and metally strong. (Hence, I'll have to get an axe at one point, just for the fun of it

As stated before, voice recognition is also a really hard thing to do, not an easy problem to solve. There is a reason why even Siri gets sent to the Internet.

Btw, as you already have a RFXLAN and now that new RFXtrx thingy, what part of the 433mhz will you let handle by the Homey, or will you make the RFXLANs sent their output to the Homey so that it can do all the centralized things it needs?
The advantage of having multiple 433mhz receivers is ofcourse that one can spread them throughout a house and avoid distance/wall issues. for reception.

Homey will handle none of that. My own system is and will stay the 'core'. (where there is no 'core' actually, it's just a bunch (>30 or so) of processes producing and/or listening to MQTT messages).
I hope I'll be able to use Homey as an additional interface that supports speech, but I don't want it to also read sensors or control lights or whatever.
I just want it to send me (ehh, my HA system) what words were recognized - my system will take care of the rest.

I would in that case suggest a NUC or some other small-form-factor, low-energy-using, PC kind of box. RPi is too light for processing voice.
But in that case you'll have to do it yourself, which is likely too if Google shutters the service or rare limits queries.

"Homey, sexy lights" will then turn in to the homey from outside coming in your house and turning your lights red instead

(which is why "homey" is also a bad name, as that is the gangster friend strange, you would htink people would have caught that when they picked the name of the toy)

PS
Regarding multiple RF receivers. I have a single driver that connects to multiple RFXCOM receivers (2, soon 3) and this driver collects all the information those receivers have to offer.
All that info is filtered on duplicates and only changes will result in messages to my MQTT broker.
So in fact it doesn't matter whether I have 1 or 20 receivers - the load on the output side of the driver (= MQTT messages) stays almost the same.

[/quote]

While you can do it that way (ignore the same message when received during a short timeframe) one can then not use repeated keypresses on a remote for instance (volume up, channel down etc). That becomes a design restriction then though.

Thinking about it though, one typically does not have many devices, except for IR, that have that issue though, everything else is mostly single press.

lets see and wait what there product will be next year, i think it is to soon to evaluate a product which is in an embryal state know and will be released in 9 months. the human mind is full of fantasy if there are no hard data avalaible, like a product . that phenomenen is called reality testing in psychological terms. Is like the wether forecast . How do you test the future. That is only possible in the mind of humans as fantasy.

labium wrote:lets see and wait what there product will be next year, i think it is to soon to evaluate a product which is in an embryal state know and will be released in 9 months

Agreed, though with the sparse info we have now, especially the important factor: Voice Control, it does not seem that will last the next 9 months unless they get absorbed in Google (which is a likely exit plan they have) to be able to offer that service (and for the big G to get a microphone in each house).

Was anyone able to install Jasper successfully?
I have tried a couple of times but the installation procedure is not plug-and-play.. There's many manual steps and dependencies. The "pre-build" version appears to be gone.
DJ

As it's now exactly 1 year later since the last post, I'm curious if anybody has any Homey experiences to share?
Is this thing as good as advertised?

Got some updates by mail but did not receive anything yet.

DJF3 wrote:Was anyone able to install Jasper successfully?

I've had it running about a year ago or so; disappointed about the results so I moved on.

BTW I read a (commercial) article on DFT this evening about "MyHomeBox", presenting itself on the upcoming IFA Berlin, claiming to be better than its competitor 'the Homey'.
How hard can that be... "we're better than our still non-existing competitor"

Two things:
http://www.amazon.com/Amazon-SK705DI-Echo/dp/B00X4WHP5E

Yep, that does exactly and way more than Homey does, and well, Amazon is a bulti-million company

Google is getting in the game:
https://on.google.com/hub/

They just forgot the microphone (for the rest it is a device you really do not want in your network due to them controlling it and doing network traces, that is... if you want a little bit of privacy left)

Both are (voice) remotes.

Digit wrote:Both are (voice) remotes.

... for Home Automation:
http://www.amazon.com/home-automation-s ... 6563140011

and this little 100 million Alexa Fund for Voice control:
http://phx.corporate-ir.net/phoenix.zht ... ID=2062558

see also:
http://www.zdnet.com/article/amazon-ech ... ot-circus/


The Google OnHub thing does not have a microphone thus no Voice input in this edition, it does have all the right chips for general remote control though.

Digit, can you please elaborate why you were disappointed?

olof wrote:As it's now exactly 1 year later since the last post, I'm curious if anybody has any Homey experiences to share?
Is this thing as good as advertised?

I suggested to take a new name for the project:

"Slomey"