Open mrichar1 opened 6 months ago
I've seen this as well after implementing a custom wake word. It doesn't trigger the wake word indefinitely but it does trigger it a second time sometimes.
I have exactly the same issue with the same log:
DEBUG:root:Wake word detection occurred during refractory period
The wakeword has been made with Google collaborate. I have different wakwords generated and the issue happens on all custom wakewords.
Any help on that would be great, because I want to use the custom wakeword.
same issue as the above users, custom wake words triggers repeatedly on Wyoming satellite. I have had it trigger 4 times after an initial trigger, stopped only by saying 'Nevermind' in the end.
I wasn't 100% sure where this issue lies, so apologies if there's a better place for this..
I have
wyoming-satellite
working withwyoming-openwakeword
in docker containers along with Home Assistant. This all works fine using any of the 'builtin' models, or with some of the models I have downloaded from thehome-assistant-wakewords-collection
repo.I built my own model using the simple model trainer on colab, and this seems to test ok for matching with
openwakeword
in theirdetect_microphone.py
test script.However when I load this custom model on my server, I get strange behaviour, in that it wakes up 'correctly' the first time, but then treats any voice it heard after that as another valid wake-word. If the microphone can 'hear' the speaker then this can lead to long loops of wake/command, unless a long refractory period is set.
Some debug logs to show what the apps are doing:
wyoming-satellite starts up (to show config options):
When a wake-word is spoken, we get the following (interspersed satellite and openwakeword logs):
Any suggestions on where to start debugging this appreciated - I can provide the custom model if helpful for testing.