Consecutive wakeword

Hey everyone!
I’d like to know whether it’s possible to listen for another wakeword after the ‘standard’ wakeword was recognized, while still also doing the stt/asr/nlu part for intent recognition. Then, when the second wakeword is recognized an event is triggered instead of any intent that might have been made up from the audio.

The reason for this is that i realized that wakewords are recognized way better than intents (makes sense, since they are specifically trained for just that word) and I’d like to have one very secure intent that gets recognized with almost no errors - hence the second wakeword.

My setup at the moment is a Raspberry Pi 4, running rhasspy as a service, appdaemon for handling mqtt messages on the same Pi and home-assistant on another pi for handling intents related to other devices. My audio input is from a matrix voice (-> ALSA).

I was thinking about starting a script in Appdaemon on asr/startListening, maybe triggering the local instance of mycroft-precise (currently only running in a venv, when I train my wakewords). But I have no clue how to retrieve the audio stream from rhasspy and use it for precise… I guess I have to use it via Rhasspy, since the microphone gets blocked from it.
As a workaround I might be able to use audioCaptured after the intent recognition is done… This would’nt be ideal since it means i lose some time before the wakeword can be recognized, but better than nothing…
I just don’t know how or where to set sendAudioCaptured in the /hermes/asr/startListening to true.

I hope you can understand what I’m trying to do :sweat_smile:
Maybe someone already has a solution for this or can give me some ideas as to how this could word?

Any help would be greatly appreciated! :smiling_face:

I’m now playing around with Node-RED instead of Appdaemon, but I still need to find a way to evoke that second wakeword… Maybe even do the wakeword detection in Node-RED instead of rhasspy to begin with.

If anyone hast auch a Setup, please let me know