On some requests, the awake-wav trigger sound plays again after the done-wav sound, then OWW gets stuck

[daytonturner]

I'm noticing that occasionally, I'll hear the wakeword trigger awake sound play a second time. After saying the wake word, I'll hear the awake trigger sound, speak my question, and hear the done sound. Then, immediately after, I hear the awake trigger sound play again. Then I get my answer, and from that point on the OWW wyoming satellite seems stuck. If i go into home assistant and tweak any of the mic settings (auto gain, mic level, noise suppression level) it will then play the trigger sound again and get unstuck.

Anecdotally, it seems to be on longer requests, or ones where ive asked two questions such as "What is the pool temperature, and what is the temperature outside"

I'm not sure what's causing this, or how I could debug it. Playing back my recordings of the command it heard me say, the mic recording seems clean and clear.

Any suggestions?

[daytonturner]

For what its worth, even though a reboot didnt seem to help, I restarted only the wyoming-openwakeword service and it seems to have stopped doing it for now.

[spdustin]

That didn't help for me. Once it starts playing back TTS data from Home Assistant, the debug log shows the microphone immediately unmuting, and the lights turn yellow on my hat.

EDIT: This only happens on custom wake words. It doesn't happen using ok_nabu, for example.

[ALX-TH]

the same here i have ... @spdustin thanks for suggestion, using built in wake words seems works almost as expected

[dreed47]

I have the same experience... none of the custom wakewords work

[s-heppner]

I've got it to work via the comments from #217. I had to copy the custom wake word model to the default model location: <dir>/wyoming_openwakeword/models (See this comment).
In case of the docker container, this was located at: /usr/src/.venv/lib/python3.11/site-packages/wyoming_openwakeword/models/.
This is also the directory I set as the custom model directory.

Then, I had to rename my custom wake word by adding _v0.1, so from example.tflite to example_v0.1.tflite (see this comment).
I have not the slightest idea of why this would work, but it does.

Having to copy a file into a Python package installed in a virtual environment is obviously an ugly workaround, though.

From my testing, I think there's also a second bug, that makes the wyoming-speaker docker container crash, if the --done-wav is still playing when the satellite wants to play the TTS response, that may or may not have anything to do with this problem in the broader sense.