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.

OK, Homey, what version are you running?

"I am currently running version undefined."

How informative.

Comments

  • Looks like they are fixing this bug in 0.8.31 (source: github)
  • Do you know where i can find a list of the standard voice comands homey already has? Like this one.
  • I just took a random guess. But, there is a topic here on the forum that tries to list a bunch of commands. Look for it. It's not very old, it's quite recently been updated.
  • Nice! Thanks!
  • I can confirm this is fixed in 8.32. My homey just responded correctly to this command.
  • Ya. "I am currently running version 0.8.32, I think. But I might be mistaken, since this is beta software."

    Cute.
  • What's the trigger for this in Dutch?
  • fiekfiek Member
    Fire69 said:
    What's the trigger for this in Dutch?
    Wat is je versie worked for me
  • Great. You know I'm really glad I did a full recovery. Even with the tv on Homey only responds to me saying "OK Homey" and she understands me (en again, even though the tv is running). I'd suggest @Athom deletes everybody's trained voice in the next upgrade. The experience will be much better. Did you have similar experiences @Fire69 ?
  • Yes, after the reset no more false positives!  But still horrible speech recognition... 
    Is your Homey set to Dutch or English?
  • Dutch...
  • Fire69Fire69 Member
    edited April 2016
    Just switched to English to test, but it's just as bad.
    I have a flow <good night>, I think Homey's mocking me.  I said 'good night', he hears 'good morning'...   :s
  • To be honest, I think the bad speech recognition isn't as much Homey itself, but is more an issue with the speech-to-text server that is used by Homey. I have to speak very clearly, put lots of space between words, and use short sentences (~2 words) and even then 2 out of 3 sentences are misinterpreted. I wish we could see which speech-to-text server Homey used, and influence this ourselves, so we could choose the one that works best for our personal situation.
  • It's possible these speech recognition servers all run the same software, so it's just a matter of load balancing here. But I could be wrong.
  • novaflash said:
    It's possible these speech recognition servers all run the same software, so it's just a matter of load balancing here. But I could be wrong.
    Yeah, I think you're wrong there. From the blog post:
    This server then forwards the audio to 3rd party services, and when they come back, we compare them to pick the best result. A few factors are weighted here such as speed, accuracy and ‘dynamic-ness’. Some servers are very good at dynamic words such as artists or movie names, but they are generally slower. So when we detect words that hint on dynamic words, such as ‘music’, ‘movie’, ‘play’ etc, the Speech to Text service at least waits for the 3rd party service that’s best at detecting those to finish. That way, we guarantee the best results per scenario.

  • In that case I really want the option to manually choose the recognition server.
  • I'm also experiencing a very, very bad speech recognition in English as well as in Dutch. My experience is that it only works when pronunciation, volume and distance to Homey are in perfect conditions for Homey and that's not human-friendly.

    Google's speech recognition on my Android phone is much, much better. The latter gets me right far more than Homey. 

    Let's hope it will improve a lot the upcoming period, because it's a must-have with speech as a key selling point.
  • What I need when dealing with Homey's speech recognition is the ability to specify:

    When: I said: "be quiet"
    or When: I said: "3 quite"
    or When: I said: "3 quiet"

    Then: set volume 30%.

    But the 'or' function is not possible. The above of course is necessary to account for the variations that Homey has when recognizing speech. The above are actual examples of what Homey thought I said when I said "be quiet". If I want Homey to get it right I have to say it like "beeeeee qui-it" like I'm talking to a moron. The above 'or' solution could help me avoid talking to the Homey like it's got a hearing problem.
  • thayoung1 said:
    I'm also experiencing a very, very bad speech recognition in English as well as in Dutch. My experience is that it only works when pronunciation, volume and distance to Homey are in perfect conditions for Homey and that's not human-friendly.

    Google's speech recognition on my Android phone is much, much better. The latter gets me right far more than Homey. 

    Let's hope it will improve a lot the upcoming period, because it's a must-have with speech as a key selling point.
    Question: Does that recognition still work perfect if you put your phone/android device 4 meters away from you ? Or are you holding the device next to you while speaking? Because distance and such greatly influence the recognition
  • The mic of your phone is not designed for voice at several meters away.
  • @jjtbsomhorst Speaking to my phone from several meters away doesn't work. That doesn't bother me. When using my phone it's always within arm range. 

    The Use Case of the Homey is a different one. I don't carry my Homey with me so that device should work with a distance. 

    What I was trying to say with the phone is that it understands me much better on short distance than my Homey does on the same short distance. Hope this will get better soon :smiley: 

  • BumblezBumblez Member
    edited April 2016
    novaflash said:
    In that case I really want the option to manually choose the recognition server.
    Looks like that's not going to happen... see Emile's reply here

    @Emile - in regard to building my own speech recognition... is there any way to obtain the raw speech data (voice recording) from Homey's microphones?
  • EmileEmile Administrator, Athom
    No, but you can send a text string to Homey from e.g. a raspberry Pi and a microphone. 

    Try using longer sentences, they usually work better because the speech parser has more context to guess what you meant. For example, not 'light 50%' but 'turn the lights to 50% brightness'. 
  • Well, that would kind of defeat the purpose... I wanted to be able to test if Homey's setup (with Homey's microphones) would work better with a different speech-to-text server, not completely circumvent Homey's setup.
  • Emile said:
    No, but you can send a text string to Homey from e.g. a raspberry Pi and a microphone. 

    Try using longer sentences, they usually work better because the speech parser has more context to guess what you meant. For example, not 'light 50%' but 'turn the lights to 50% brightness'. 
    Weird, because voice commands like "radio aan", " TV aan" have a 100% hit ratio
  • Following @Emile's advice, I've been doing a couple of test runs, and I have to say, this is very useful advice. For example:
    "TV on" has a recognition rate of approximately 33%, whereas "turn on the TV" is recognized correctly almost 100% of the time. So I'm going to update my flows accordingly...
  • ARJARJ Member
    @Bumblez What kind of keywords are you using in the flow "turn on the TV"?
  • BumblezBumblez Member
    edited April 2016
    My flow there is:
    If [you said anything]  and [text contains "turn on"] and [text contains "TV'] then Harmony:[Start activity {TV}]

  • ARJARJ Member
    edited April 2016
    Thanks
Sign In or Register to comment.