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.
The Homey Community has been moved to https://community.athom.com.
This forum is now read-only for archive purposes.
Maybe Homey needs a hearing aid?
Spawnreaper
Member
I am getting slightly frustrating. I don't even want to show of homeys capabilities to friends because 8 out of 10 times voice commands do not work.
Now I've been checking the speech input logs and homey hears less clearly then my grandmother (93).
now the only thing i am screaming to my homey is "TV ON!".
Eventually I just input the command manually for it to work.
I wish my experience was better, but its getting very frustrating.
Now I've been checking the speech input logs and homey hears less clearly then my grandmother (93).
now the only thing i am screaming to my homey is "TV ON!".
Eventually I just input the command manually for it to work.
I wish my experience was better, but its getting very frustrating.
Comments
So use " When I say something" and the logic card containing the key words in the and part of the flow?
I also use the function when homey listens to mute my receiver when I say ok Homey.
I get a succesrate of about 80% depending on the range from Homey.
There are already many topics about speech in this forum. It might be better to react on an existing one so that the forum stays compact.
So my wife took the Hue app and fooled/tricked us......
After the friends left I took a look at speech-input and I was amazed; Homey triggered that evening >20 times while we sat 5-6 meter away from it.
So the hearing is good, but the understanding is terrible.
That said, yes, voice recognition on Homey is still a big concern... and Athom are well aware of it and are actively working on it. See also the remark on the June 3 Software Status update about them playing around with Google's Speech API...
Disadvantage of large sentences is that everyone at home should say that exact sentence, so that is not very handy.
We "have" scenes for dinner (word: eten) or cosy (word:gezellig) and sometimes they work, but mostly not. My kids now already scream to homey in the hope it will listen.
So i did some testing and put a starting sound when homy starts listening and a end sound when homey stops listening. What i found out is strange. When i say for example: "zet de lampen gezellig" it fires the sound of stop-listening when i'am still talking. That happens many times, so large sentences do not work for me that great.
But giving commands like "eten, gezellig, tv on" is not the way i would like my kids to talk to people... or devices.
So in my flow homey listens to the word "gezellig", (you said something + contains "gezellig") but i try to call him bij saying "maak het gezellig" or something like that. And yes, pure the word "gezellig" also does not work very often anymore. In the beginning is worked better but with more apps and flows it got worse.
So i hope Athom wil come with a new release very soon for the people who are not in the experimental channel, because at home Homey is loosing his cool status.
If [You said anything] AND [text contains "turn on"] AND [text contains "TV"] THEN [start Activity "TV"], [Say something: "Turning on TV"]
So as long as you agree on the words to use, the exact order of them in the sentence doesn't matter (in my case, only the words "turn on" have to be in order, as separate statements for "turn" and "on" are too error prone).
As for a new release... don't expect any big improvements to voice recognition when the 0.8.38/0.8.39 goes from experimental to release status... We'll have to wait a while longer and give Athom more time, before we can expect any changes there.
I also realised that this can not be used since the actual voice recognition is processed online and before that its just an unusable audio file.
Although it has benefits using the processing power of a more powerful server online the down side will always be that it does not send the audio one word at a time therefore it will never know if your sentence is finished. Its also not the servers fault because it can only process audio it receives.
I think it is fair to say that we have to wait for technology to improve so it can handle voice recognition and language interpretation locally.
That said, I do have some ideas to improve what we have by adding something to a flow.
"Estimated use period"
U usually do not turn on the lights in the daytime so maybe adding something that allowed homey to estimate what you probably would say can help it to trigger the right flow?
just an idea?
Why would it not send one word at a time?
Detecting a pause and sending the 'word' while listening to the rest is possible. (That's what Siri does)
Obviously it would require an extra step on Homey to construct the full sentence off all collected words, but that may be faster then listening for a full sentence and then sending it and waiting for a result.
And as a next step of improvement it could send that sentence to a webservice for correction (That's what google search does when you type a crappy sentence)
And if speech to text conversion is not entirely accurate, then sending the results to another web service for correction would most likely turn it into something completely different (apart from the fact that it would introduce another delay in the command processing, as Homey would have to wait for the results from that web service as well). Personally, I would very much prefer my Homey to NOT use a speech correction service on the sentences I speak.
Even in google search, more often than not, the "did you mean ..." is absolutely not what I want.
Same problem here, i can not use homey because 9 times out of 10, dont understand me. I hope athom knows this problem with the speech input.