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.

Is there A problem with the Homey microphone?

From the beginning I had a lot of problems when speaking to my new friend Homey. Almost nothing was exepted. Not in Dutch neither in English. 

Now I have the app on my IPhone and it almost works flawless. So I am wondering, is they an issue with the microphone of the Homey. This since speech recognition works on the iPhone but not when talking directly to Homey. 

There is one exception and that is when I am very close to Homey.  Than it works, not always but better. But when further away than 50 cm than it really sucks. 

Would appreciate to receive some feedback from the developers if this an issue under construction or is it staying like this. Don't want to use my phone when sitting in the same room with my new friend. 
«1

Comments

  • PilsPils Member
    For more or less the same experience. Talking to the new iphone app works very good now. But talking to homey itself almost never works anymore. But, a few months ago, it worked better. Also my kids turned on the lights by saying: "OK Homey Eten". Now they don't even bother because Homey never 'listens'.

    So the good news is, i think it is not the hardware (alone), because it has worked better than it does now. But the sad news is, that it never worked good enough to be used in real life.

    Also i noticed a bug when i was testing this. I mean the ios app and homey itself. When talking to app fist, it responds by talking to my iphone. But when i say "ok homey" to homey (from very close), it responds thru my iphone instead of thru it's own speaker. Also when not using the app anymore. 
  • mglatmglat Member
    Hi Arthur, go to http://yourhomeyip/manager/speech-input and see what Homey hears when you talk. 
    There is also a learning option to get homey better understand your voice. I didn't check that out yet so don't know if that will work.

    Also splitting the flow tends to do better. So in stead of the <you said something>"...." change it to <you said anything> and <it contains>"..." and <it contains>"...."
  • BumblezBumblez Member
    edited July 2016
    It's not the microphone. Hardware wise, the two microphones that are in Homey are actually pretty good ones. It's the speech recognition intelligence (translation from speech to text) that is behind it, that is the problem. Athom are aware of this and they are experimenting with different solutions (such as using Google's speech API, the same one that is also used on Android phones). So we will just have to wait for the solution (don't expect anything on a very short term, though).

    The good news is, this can (most likely) be fixed with a firmware update
  • jjtbsomhorstjjtbsomhorst Member
    edited July 2016
    Bumblez said:
    It's not the microphone. Hardware wise, the two microphones that are in Homey are actually pretty good ones. It's the speech recognition intelligence (translation from speech to text) that is behind it, that is the problem. Athom are aware of this and they are experimenting with different solutions (such as using Google's speech API, the same one that is also used on Android phones). So we will just have to wait for the solution (don't expect anything on a very short term, though).

    The good news is, this can (most likely) be fixed with a firmware update
    I'm sorry to say but that is not correct in my opinion. How come Homey understands me perfectly when talking to it through my phone and it does not when talking directly to him? That is because the internal microphone is not that good adjusted as the ones found in your mobile devices. Another reason is that talking to your phone, with a distance around 30 cm's does not suffer that much from environment then Homey does when you talk to it 1 or 4 meters away. 

    So personally I don't think they can solve it for everyone. I would suggest that everyone is talking to homey using their phone and have a great working speech recognition instead of trying to command your homey without and have a big fail rate. 
  • BumblezBumblez Member
    edited July 2016
    You're wrong on the first part. When talking to (the Homey app on) your phone, you use the speech-to-text service of the phone itself (either Google's speech API, or Apple's speech API), not the speech-to-text service of your Homey. The app uses the speech recognition options that are built into your phone. So the phone sends the interpreted text to your Homey, not the raw audio file.

    But you're right about the second part... when you talk into your phone, the distance to the microphone is  lot less, which definitely contributes to the fact why voice recognition of your phone is a lot better.

  • Bumblez said:
    It's not the microphone. Hardware wise, the two microphones that are in Homey are actually pretty good ones. It's the speech recognition intelligence (translation from speech to text) that is behind it, that is the problem. Athom are aware of this and they are experimenting with different solutions (such as using Google's speech API, the same one that is also used on Android phones). So we will just have to wait for the solution (don't expect anything on a very short term, though).

    The good news is, this can (most likely) be fixed with a firmware update
    I'm sorry to say but that is not correct in my opinion. How come Homey understands me perfectly when talking to it through my phone and it does not when talking directly to him? That is because the internal microphone is not that good adjusted as the ones found in your mobile devices. Another reason is that talking to your phone, with a distance around 30 cm's does not suffer that much from environment then Homey does when you talk to it 1 or 4 meters away. 

    So personally I don't think they can solve it for everyone. I would suggest that everyone is talking to homey using their phone and have a great working speech recognition instead of trying to command your homey without and have a big fail rate. 
    There is a lot of variation. For me its quite oke, I even talk to Homey from 7 meters away, and its going wel, I have to change my voice a little and talk softer if I'm sitting close to it. and during the day the voice to Speech is not consistend, but moost of the times it works good. 

    And allways talking to your phone is not an option for me, as i'm happy to put my phone down once in a while, and if it's charging ... 

    Lets see what "Improvments" can do and than everyone can choose if the will there phone as a primairy ,
  • AcE_KrystalAcE_Krystal Member
    edited July 2016
    I'm having kinda the same feeling about the less good mic in homey, and also found talking to my phone (Samsung Galaxy S4) works way better. I'm unsure about what translator being used when speaking to your phone. I thought it was all the same translator, but I'm unsure now.

    So it may be true that a big portion of the problems is in the translator. 

    Disclosure: The following are my personal toughts XD
    Though I don't really thing the problem is in the mic hardware itself, but rather a not so optimal implementation of the mic inside homey. I read they actually spent some good care in selecting a high quality mic. But Mister High Quality Mic might suffer a bit from sitting inside a big echo shell without sound isolation together with 2 new rather noisy friends,  Mr and Ms springs (antenna's, you can actually hear the springs resonating when homey speaks).

    A other thing is that I personally found homey to listening a bit better when standing on top of something soft/sound absorbing.
    And a room with lots more echo seems to make it worse for me. The homey are our office is much harder to become friends with then at home in my super sound absorbing ger/yurt tent.

    Edit:
    One thing I DO find improved though, and I would like to mention this, is the "Ok Homey" trigger.
    Both our homey's seem to get almost none false triggers anymore compared to the past where I would get one almost every 2 sentences I spoke to someone. Playing music used to be almost none stop triggers.
  • I think they have the noise ( drempel ) level set higher,
    I have to yell at HOMEY before he hears what.

  • MarcoFMarcoF Member
    Emile confirmed there's a lot tweak able and work to be done on making the mics better. 

    Athom said this month's ago and in the mean time I didn't see any big improvement on the mics/speech recognition. Maybe better said: there's room for much more improvement.

    https://www.kickstarter.com/projects/athom/homey-the-living-room-talk-to-your-home/posts/1523076
  • I can't believe anyone has tested this yet, but I went and compared Homey's microphone and interpretation with my phone's (LG X screen). I placed my phone next to my Homey and yelled different commands in the room.

    Results in the form of:
    What I said
    homey: What homey heard
    mobileXXXX: What the phone heard

    is it going to rain
    mobile23786: is it going to rain
    homey: is it going to rain

    what are you
    homey: what are you
    mobile37903: what are you

    What is the temperature at this moment
    mobile48927: what is the temperature of this
    homey: what is the temperature of kitchen

    I have a question about thermodynamics
    homey: i have a question should we arrange
    mobile63242: i have a question about the

    Turn on the livingroom lights
    homey: turn on the lights
    mobile54164: turn on living room light

    Turn off the livingroom lights
    homey: turn off the lights
    mobile32513: turn on living room light

    What is your purpose
    homey: what is your wish
    mobile21071: what is

    Is it going to rain in the next 30 minutes
    homey: is it going to rain in manchester this
    mobile7196: is it going to rain in manchester

    What is 22 times 23
    mobile20243: what s 22 times 3
    homey: what is 22 times 3

    Small test, but in conclusion: there is absolutely no difference if you take out the distance variable between your phone and your homey. These samples were spoken around 3 meters from Homey/phone away. I do not think homey and phone use different text2speech interpreters, nor is Homey's microphone significantly worse or better than my phone's.
  • supericesuperice Member
    edited July 2016
    Bumblez said:
    You're wrong on the first part. When talking to (the Homey app on) your phone, you use the speech-to-text service of the phone itself (either Google's speech API, or Apple's speech API), not the speech-to-text service of your Homey. The app uses the speech recognition options that are built into your phone. So the phone sends the interpreted text to your Homey, not the raw audio file.

    But you're right about the second part... when you talk into your phone, the distance to the microphone is  lot less, which definitely contributes to the fact why voice recognition of your phone is a lot better.

    That's not entirely true actually, the phone uses the same speech-to-text service and detection libraries as Homey itself. The trigger detection library is the same (the 'OK Homey' part), and it uses the same speech-to-text conversion servers as well. The app records audio, sends it to our processing servers, gets the text result back, and sends that to Homey.

    That second part is much more important: when talking to your phone you always have it close to your mouth, which makes speech-to-text dead simple.

    Hearing a person from across the room (especially with background noise and/or echoes because of the walls) is much more complicated. Close-by these side effects are neglectable, but when trying to record a person talking across the room eliminating those other sounds is much more important. As you can imagine, most speech-to-text services aren't being configured get that right, because there aren't that many speech-to-text products that try to understand people from across the room, and speech-to-text is hard enough as it is. We're trying to improve the situation by using a dedicated preprocessing chip which can do some pretty neat stuff, like microphone beam forming and noise cancellation, but of course the configuration of that chip can always be improved.
  • BumblezBumblez Member
    edited July 2016
    superice said:
    That's not entirely true actually, the phone uses the same speech-to-text service and detection libraries as Homey itself. The trigger detection library is the same (the 'OK Homey' part), and it uses the same speech-to-text conversion servers as well. The app records audio, sends it to our processing servers, gets the text result back, and sends that to Homey.
    Thanks for the info. I definitely did not know (or even expect) that. But if that is the case, then what exactly did Emile mean
    when he said
    "We've also played with Google's Speech API. It looks promising to increase the Text to Speech accuracy, but we have to perform more tests to be sure about that!"
    because I thought this was about replacing Homey's current speech-to-text functionality with the Google API's speech-to-text functionality (and the latter should already be what is used on Android phones, right?)
    And yeah, I do assume the "Text to Speech" in his sentence was supposed to be "Speech to Text", because text-to-speech is not an issue.
  • supericesuperice Member
    edited July 2016
    Bumblez said:
    [snip] But if that is the case, then what exactly did Emile mean 
    when he said
    "We've also played with Google's Speech API. It looks promising to increase the Text to Speech accuracy, but we have to perform more tests to be sure about that!"
    because I thought this was about replacing Homey's current speech-to-text functionality with the Google API's speech-to-text functionality (and the latter should already be what is used on Android phones, right?)
    And yeah, I do assume the "Text to Speech" in his sentence was supposed to be "Speech to Text", because text-to-speech is not an issue.
    Replacing the backing service of our speech-to-text servers. We've abstracted speech-to-text to go through our own servers, so we can choose serverside whether we'd want to use a third-party service, create our own, or run recognition software on our servers.

    We have a number of reasons to do that, like being able to switch speech-to-text providers very fast, comparing the results of different stt providers and/or postprocessing the text (e.g. in a context of lights, 'red' is much more likely to occur than 'read', so we could replace that on our side already). In this case Emile was talking about switching our servers to use Google's Speech API, which might yield better results in some cases.
  • ok, thanks a lot for the clarification... I really appreciate it.
  • It seems to work better now after the update
  • WathLamersWathLamers Member
    edited September 2016
    Arthur27 said:
    It seems to work better now after the update
    what update?

    edit   for me the recognicion is bad.
    homey does not understand me at 2,5 meters.   only if i shout he listen.
    oke homey   however works fine.
     whit version 0.9.1     the recognicion was fine for me...
    But  one way or the other...    nobody at this forum had the same experiance...
    It is bad this way......   very  very bad
  • My Homey has been extremely deaf for months.  I had to talk to it at max 50cm or he wouldn't understand anything.
    It was located on a small table against the wall next to my couch, a bit cornered.
    A couple of days ago, as a test, I moved it to the other side of the room, same height against a wall but nothing to the left or right.
    It's still not good, but the recognition is much, much beter, even when talking from further away (not far, 1 or 2m) !
    Also, the orange rings now 'responds' a lot better to my voice, like the volume is higher although I talk from further away.

    So I don't know the details of the microphones, but the location of your Homey seems to be very important...
    I think they need to do a lot of work on the software for the microphones, noise cancelling and stuff...
  • Arthur27 said:
    It seems to work better now after the update
    what update?

    edit   for me the recognicion is bad.
    homey does not understand me at 2,5 meters.   only if i shout he listen.
    oke homey   however works fine.
     whit version 0.9.1     the recognicion was fine for me...
    But  one way or the other...    nobody at this forum had the same experiance...
    It is bad this way......   very  very bad
    Same here, the speech recognicion is a big problem!
  • Same problem here from the start. No progress noticed since april.

    Love these videos of Alexa which looks like working better (haven't seen / experienced it live but I bet user experience with speech is better).
  • @thayoung1 ;
    Please be careful with these movies. We have seen movies of Homey with perfect speech, KaKu indoor ranges of 30m+ through thick concrete walls, read stories about perfectly working presences detection, brilliantly working IR,  etc 

    So please be careful with movies of companies which want you to sell something ;)
  • thayoung1 said:
    Same problem here from the start. No progress noticed since april.

    Love these videos of Alexa which looks like working better (haven't seen / experienced it live but I bet user experience with speech is better).
    Assumptions Assumptions...
  • olive said:
    Arthur27 said:
    It seems to work better now after the update
    what update?

    edit   for me the recognicion is bad.
    homey does not understand me at 2,5 meters.   only if i shout he listen.
    oke homey   however works fine.
     whit version 0.9.1     the recognicion was fine for me...
    But  one way or the other...    nobody at this forum had the same experiance...
    It is bad this way......   very  very bad
    Same here, the speech recognicion is a big problem!
    The recognition is not the problem...
    Try talking to it from max 50cm away, Homey will understand you almost perfectly. 
    The problem is that it doesn't seem to capture the sound from a distance.
    And only for certain people, others have no problems at all with it. 
  • ZperXZperX Member
    edited September 2016
    Recognition is a problem. I don`t know how is dutch but English is terrible. My pronunciation is poor but and when I speak slowly and emphasising word I have 1:3 success rate. When my native speaking (UK) relatives or friends try it they have no chance at all, no matter of the distance.
    Should switch to google or amazon recognition.
  • Dutch and English work 'perfectly' here.
    Try using the app and check the /speech-input page.  If recognition is OK there, the problem is with the microphones, otherwise it's your pronunciation.  Maybe your relatives' English is too 'flat' :p
  • Did you try speech recognition with the app on iOS or Android? Than you can conclude if it's the microphone or your speech. 
  • Fire69 said:
    Maybe your relatives' English is too 'flat' :p
    The technology should serve us not us serving the technology. Yes I am very familiar with the speech input page. I was thinking to make a video and post it on youtube to show how ridiculous the interpretation of simple commands are. But I refrained doing it, don’t want to make bad marketing for Athom. But if they won`t tackle this issue for long I might do it to demonstrate the depth of the issue. It feels like the speech recognition 7 years ago. If the Cambridge accent is not good enough than I don`t know what to say. Shall we speak Geordie, Scottish or with Alabama accent?

    Oh yes there is no issue with the app, that works perfectly like it should. I reckon that uses the google speech to text engine.

  • ZperX said:
    Fire69 said:
    Maybe your relatives' English is too 'flat' :p

    Oh yes there is no issue with the app, that works perfectly like it should. I reckon that uses the google speech to text engine.

    Nope, same speech engine, so that proves enough I guess  ;)
  • How do you know it is the same engine?
  • ZperX said:
    How do you know it is the same engine?
    Superice confirmed it in the Android-app topic.  I'm searching for it, but with the crappy search it's no easy task  :|
Sign In or Register to comment.