This is the forum archive of Homey. For more information about Homey, visit the Official Homey website.

The Homey Community has been moved to https://community.athom.com.

This forum is now read-only for archive purposes.
Closed

Kickstarter vs Current state.

Back in early 2014 when browsing kickstarter there was this campaign that caught my eye. 
This lovely round white ball, with a beautiful LED ring spinning in thousands of colors. 
Like all other campaigns, this campaign contained several videos showcasing the product. Emile and Stefan very passionately described how this tiny white ball would change the way I interacted with my TV, lights, plex, central heating, etc. 

I was sold right away there and then. I grabbed my monitor and shouted very loudly "shut up and take my money!", of course that didn't work. Homey had not yet arrived so I had to go about this the old fashioned way, grab my mouse/keyboard and pull out my credit card. After checking with the boss (a.k.a. "the wife") for approval I decided on purchasing the geek edition of the product, being a big bald IT nerd that really was the only way to go. 

After pushing the "pledge" button in exhilaration and signing off my life savings I prepared myself for the next inevitable phase in the campaign. Any and every kickstarter campaign has this phase, it is called "The Big Wait". (it is like the time between takeoff and landing. no fun, just sit back, relax and have happy thoughts about your destination)
During this period Emile and Stefan kept us eager Geeks up to date with posts, videos, facebook updates and whatnot... 

I got to celebrate a new year, and a year later I got to celebrate one more. But shortly after the pretty fireworks had dissipated an email arrived. It contained the first and best good news of the year. My homey had been shipped! 

Carefully opening the package I pulled out the, bigger then expected, white ball of happiness.
Plugged in the USB cable and put it down on my TV stand. Unfortunately due to a rugged high quality cable USB that didn't bend very well I almost got to see how good homey was at bowling but some quick reflexes saved the day. 

After connecting Homey to my high speed enterprise home wireless network and the ring starting to glow exactly like on the kickstarter videos I was very eager to start teaching my new toy how to control the other electronics in the house. 
Having remembered how easy it was to connect something, I grabbed the remote control for my lights and sat next homey. I said to my new found friend: "Homey, I would like to connect a new lamp please". Nothing happened. I repeated myself oncemore, still nothing happened. I double checked the videos from kickstarter:

Even it's creators own voice thundering from my speakers did not persuade homey to actually be interested in my lights in any way whatsoever.

Being an IT nerd, the one thing I didn't do was to read the f...in manual. I then learned I got it all wrong. Renewed insights from its creators meant the activation phrase was now "OK Homey". Wow that was easy!

"OK Homey, I would like to connect a new lamp please"
"OK HOMEY!?, I would like to connect a new lamp please" 
"OK HOMEY!?, I would like to connect a new lamp right now!" 
"OK Homey?"
"OK Homey!"
"OK Homey?!"

Hmmm well.. Nothing... Maybe the Chinese guy who put it together forgot to change the chinese language test chip for the english one?

I then proceeded with the voice training to make sure it would understand my "louis van gaal" English. I had always thought my English wasn't that bad. Having to read/write/speak/listen to it at work every single day I do know, unlike Louis, that soccer balls don't get players horny.
Unfortunately Homey didn't agree and if I look back at the experience he must have been stuck in torture mode. 
I literally shouted/screamed/cried/begged/whispered/yelled "OK Homey" dozens and dozens of times but it wasn't until my throat was soar that all 5 boxes of the voice training had been checked.

At that point I was a happy camper. Homey had learned how I, in my "stonecoal english" pronounce his name, or so I thought.  
It still didn't respond to my kind request to connect a new lamp, nor would it do anything else. 
The ring now did occasionally start spinning orange and white (being the sherlock holmes that I am I deducted that orange would probably mean "listening" and white would probably mean "processing"), but Homey would not do anything other then say "NO!" to me every once in a while, even if I didn't even ask him anything. 
I dropped on the couch exhausted from my struggle with the cute white evil ball. Being bald I was unable to pull out any hairs on my head, but Homey did what my 2 year old daughter had been yet unable to do: I gave up on learning it anything more that evening.

Now a wee bit more serious:
After fiddling around with the website I eventually got it to turn my lights on and off after adding the remote control as a device and creating a flow that would listen to "lights on" and "lights off". However it is still nearly impossible to get a consistent response on my voice commands. The voice recognition is *really* bad and nowhere near as refined as in the kickstarter demos. The kickstarter videos being nearly two years old, I had expected the experience and the refinement to actually be a lot better while in fact it has dropped to a level where it is nearly useless and nowhere near what those videos sold me on. Homey usually starts listening (turns orange) after I say "OK" instead of "OK Homey" (it starts to listen too soon), also it very frequently responds to ambient sound. Not only voices on TV or on the radio, but also ambient noise like a car driving through the street when my window is open. It is very frustrating that a speeding car is able to get homey to start listening and my voice actually isn't.
In case you think it might be my voice or pronunciation: Siri, Cortana, Google Search work just fine. From those services I get the recognition hit ratio I would expect for a non-native English speaker (at least 90%+). "Hey Siri" even works 100% of the time for me, where "OK Homey" works at best 10% of the time.

This post isn't intended as a flame. What you have accomplished so far has been magnificent, this product will be great, it will take a lot more time then you had originally planned but I am still confident it will be worth the wait in the end.
I obviously understand the product is still in alpha/early beta stage at the moment but I strongly feel the functionality shown in the kickstarter videos, and the videos between actual shipping should be the baseline functionality that should work out of the box right now! It has been nearly two years since the kickstarter campaign.
Go and review your past videos. Try to see and understand what expectations you have set with your very own videos and how high the bar currently is with other voice enabled products like Echo and Siri.

Please give me the smooth experience Stefan is showing off in those videos.

Keep on squashing bugs!

Thanks for reading,

Michiel

Comments

  • Nice story, I like your sense of humor  ;)

    Unfortunately, all of this has already been told several times by lots of users.
    We'll just need to have a little more patience, they're hard at work at trying to fix the problems and add more functionality.  What more can we ask from Athom? :smile: 

    Oh, by the way, "Hey Homey, I'd like to add a new lamp"?  Nice try, but I don't think they ever did that in a demo, and I doubt it will ever be possible to do it like that :smile: 
  • Really good read. You made me smile a few times and a few chuckles broke the silence around me.. Until you got to the serious part and i felt this wierd wet sensation in my eyes for a while. 

    8\10. Would buy the autobiography.

    All to be done is wait. We had a lot of practice (the backers even more) so we should be able to handle it. 
  • I already gave my opinion about the voice and you are right the other services work much better. That's where I think Athom's strategy is wrong. They have a great and ambitious idea. As a developer I know it can work eventually but there is the catch "eventually". I have worked with various speech api's and boy I can tell you it's really hard to do right. In this thing I think Athom tries to reinvent the wheel. The speech api should be much more advanced for developers. Not only that they should search for a partner that already is working on it and make that not "their" problem. They should focus on the products features because if you want to create a speech api from bottom up. You will be in for a treat.
  • JohnyTango: Good read, thanks for the humorous touch! :smiley: 

    @kitkat: you might be right there.. Although it has a huge wow-factor if it works correctly, I for one could happily do without the voice recognition and control Homey just via the app and\or a web interface, or IR.
    IMHO, Homey's Unique Selling Point is not the voice recognition but the merger of different Home Automation technologies\protocols in one device.

    Question to those lucky ones already having a Homey to play with: does the firmware allow you to skip the voice recognition training?
  • DenW said:
    JohnyTango: Good read, thanks for the humorous touch! :smiley: 

    @kitkat: you might be right there.. Although it has a huge wow-factor if it works correctly, I for one could happily do without the voice recognition and control Homey just via the app and\or a web interface, or IR.
    IMHO, Homey's Unique Selling Point is not the voice recognition but the merger of different Home Automation technologies\protocols in one device.

    Question to those lucky ones already having a Homey to play with: does the firmware allow you to skip the voice recognition training?
    Yes it's not even mandatory. You have to go to settings to do it. You can also remove voice input from there so Athom knows there will be people who will not use the voice at all. :)
  • kitkat said:
    DenW said:
    JohnyTango: Good read, thanks for the humorous touch! :smiley: 

    @kitkat: you might be right there.. Although it has a huge wow-factor if it works correctly, I for one could happily do without the voice recognition and control Homey just via the app and\or a web interface, or IR.
    IMHO, Homey's Unique Selling Point is not the voice recognition but the merger of different Home Automation technologies\protocols in one device.

    Question to those lucky ones already having a Homey to play with: does the firmware allow you to skip the voice recognition training?
    Yes it's not even mandatory. You have to go to settings to do it. You can also remove voice input from there so Athom knows there will be people who will not use the voice at all. :)
    Excellent! Thanks kitkat..
  • Fire69 said:

    Oh, by the way, "Hey Homey, I'd like to add a new lamp"?  Nice try, but I don't think they ever did that in a demo, and I doubt it will ever be possible to do it like that :smile: 


    Check the video in the post Fire69, it is from the kickstarter campaign and they show off exactly this feature :)

    I still am curious, did Athom actually look at incorporating well established voice recognition software like for instance Nuance? (Siri is/was even based on that)

    What I also forgot to mention, the current voice training option where you learn homey exactly how you pronounce "OK Homey" does not train homey to recognize your voice for other commands at all. I used dragon dictate in the past, and to get voice recognition as high as possible you could start a training mode where you would just read lots of text out loud to give the application the opportunity to learn your pronunciation for basically every word, not just the start command. Would be great if there was an extended voice learning option like that in Homey. It would certainly greatly improve its ability to understand me.



  • weird that video i have not seen a homey work like that
    was this video manipulated to show its works beter then it dos ?
  • Fire69 said:

    Oh, by the way, "Hey Homey, I'd like to add a new lamp"?  Nice try, but I don't think they ever did that in a demo, and I doubt it will ever be possible to do it like that smile 


    Check the video in the post Fire69, it is from the kickstarter campaign and they show off exactly this feature

    I still am curious, did Athom actually look at incorporating well established voice recognition software like for instance Nuance? (Siri is/was even based on that)

    What I also forgot to mention, the current voice training option where you learn homey exactly how you pronounce "OK Homey" does not train homey to recognize your voice for other commands at all. I used dragon dictate in the past, and to get voice recognition as high as possible you could start a training mode where you would just read lots of text out loud to give the application the opportunity to learn your pronunciation for basically every word, not just the start command. Would be great if there was an extended voice learning option like that in Homey. It would certainly greatly improve its ability to understand me.



    Totally forgot about that video, it's been almost 2 years since the campaign

    They are using an external company for voice recognition. Several even! Your speech gets sent to several servers and they use the response from the first one that replies. 

    Indeed, the training is only for the 'ok homey'  trigger. 
    Logical since that is the only speech that is processed locally. 

    I don't know about training... What is the use of training to your voice when you have several family members who want to talk to it? 
  • As far as I know the use only one service (Ivona) and Ivona has a cloud solution with multiple worldwide cloud endpoint which can be used. 

    Would be very strange to use 2 companies/services for one solution, this because you can guarantee the responses are the same. 

    To give an example:
    It's impossible to use Bing and DuckDuckGo if you demand the same answer to your questions every single time. These companies use different algorithms and will definitely respond with different answers to the same question. 

    Athom uses Ivona and on there site you also can listen to the same voice that are available on homey for download. 
  • Well, @emile said that's how they work several times, but it doesn't matter to the situation, they don't process the speech themselves... 
  • Where did he told it?
  • Here on the forum. I'd have to search for it, but I don't want to  ;)
  • Ah ok :smirk: 
  • Afaik Ivona isn't used for recognition, only for output... klik  :)
  • Ok,  but where's the speech to text handled? 
  • From your conversation with Emile in december 2014 : https://wit.ai/ ;)

  • ThiemenThiemen Member
    edited February 2016
    I also thought Emile stated that the request is send to multiple speech services, but when i read it again now, it is also possible that Athom just sends your request to 1 speech service, that has a few servers available, so it could still only be e.g. Ivona...

    Athom's servers forward the request to a few servers, to always get the fastest result  And our servers are redundant as well, so little chance it goes offline.
  • Fire69Fire69 Member
    edited February 2016
    Hm, wonder what happened to the Wolfram Alpha app...
    https://forum.athom.com/discussion/comment/1794/#Comment_1794
    @emile?
  • EmileEmile Administrator, Athom
    Let me clear it up for you guys:
    • Recorded speech from Homey's microphones goes to stt.athom.com, which forwards it so a few 3rd party speech processors. We compare the results for accuracy & speed, and return the best to Homey.
    • Homey talking back is software by IVONA.
    • The speech quality is not really good right now because we need a lot of voice data before we could do a good calibration. The microphones in Homey are really good and are not the problem.
    @Fire69, Wolfram Alpha was too expensive for us.
  • Thanks for the info :smile: 
  • Thanks for the info :smile: 

    Concerning the speech quality, this is only for the 'ok homey' part I guess?
    Do you have any info about the problems with triggering/training?  Is it just because the lack of voice data? Or is there actually a sofware problem?

    That's too bad about Wolfram Alpha... will there be an alternative?  Because in the other topic you said Alexa is more limited than Homey.  But without an app like Wolfram Alpha, Homey is not that 'smart'.
    We can use the Wikipedia app, but that's way more limited.
  • @Fire69 : I guess we could all install a own version of a Wolphram Alpha app with a 2000 non-commercial calls per month :)
    http://products.wolframalpha.com/api/

  •  Sounds as an creative idea ;-)
  • That would be nice indeed :smiley: 
  • Emile said:
    Let me clear it up for you guys:
    • Recorded speech from Homey's microphones goes to stt.athom.com, which forwards it so a few 3rd party speech processors. We compare the results for accuracy & speed, and return the best to Homey.
    • Homey talking back is software by IVONA.
    • The speech quality is not really good right now because we need a lot of voice data before we could do a good calibration. The microphones in Homey are really good and are not the problem.
    @Fire69, Wolfram Alpha was too expensive for us.
    Hi @Emile,

    Thanks for the elaboration. I do have some questions\remarks.
    • Is the data sent to 3rd party speech processing anonymized in any way?

    • With regard to the voice recognition and the microphone, there have been posts of users who say that voice commands over the Homey App work just fine. Others say Homey itself responds to just about every noise it hears.
      You mention that Homeys microphones are "really good" and I am wondering if they don't pick up too much information. Good microphones usually have a very broad frequency range, so there is much more information in the signal that is sent to voice processing than when using a voice microphone. The superfluous information in the signal of the really good microphones could affect the voice recognition process. A phone microphone is by design much more (but not only) catered to voice, but it does mean a lot less audio information is sent to voice recognition. That should make that process easier, faster and more reliable. You don't need really good microphones, you need the right microphones. This would explain why voice commands over phone work better than when speaking to Homey. 
      Just a thought, I may be wrong because I obviously don't know how exactly you are processing voice data.

      You also say you need a lot of voice data for that calibration. So basically, the more Homeys are out there the better they will work. I'd be happy to be of assistance in that process.. ;)
    @All
    A Wolfram App would be awesome! Great idea!
This discussion has been closed.