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.
Closed
[OK Homey] More robust trigger detection?
Hi all!
As mentioned in other topics calibrating the OK Homey trigger with more people leads to a very active Homey.
I don't know much about speech recognition but do know some signal processing basics. I imagine some kind of template matching is done in the time domain (or frequency domain of i.e. the last second of input). Calibrating your voice probably adjusts the basic template to be a better match with your voice. When doing this process with more people I would reckon the difference in voices gives the template a large spread which results in more false positives because more inputs get recognized as OK Homey.
Maybe it's an idea to add an extra calibration session for each household member (this could be coupled to the user system too) and then instead of comparing the incoming voices to one template compare them with the created multiple templates. I can't imagine this would cost too much computing power.
I want to be clear I only mean this to reduce the false positives and not as a way to have user-recognition by voice. That would need a more reliable system.
My apologies if I "sla de plank compleet mis"
As mentioned in other topics calibrating the OK Homey trigger with more people leads to a very active Homey.
I don't know much about speech recognition but do know some signal processing basics. I imagine some kind of template matching is done in the time domain (or frequency domain of i.e. the last second of input). Calibrating your voice probably adjusts the basic template to be a better match with your voice. When doing this process with more people I would reckon the difference in voices gives the template a large spread which results in more false positives because more inputs get recognized as OK Homey.
Maybe it's an idea to add an extra calibration session for each household member (this could be coupled to the user system too) and then instead of comparing the incoming voices to one template compare them with the created multiple templates. I can't imagine this would cost too much computing power.
I want to be clear I only mean this to reduce the false positives and not as a way to have user-recognition by voice. That would need a more reliable system.
My apologies if I "sla de plank compleet mis"
This discussion has been closed.
Comments