narrow mirage
Most of my personal issues (and those reviews online) using voice assistants is the required delay between the wake word and the user command.
Why couldn’t the voice satellite hardware always capture the past X seconds of audio (locally of course) so that when the wake word is detected then all the audio after wake word is sent to HA?
paper lagoon
I believe there's problem with switching the responsibility. Case is, that microwakeword and voice_assistant components are using different audio streams (in PE they actually use different microphone channels). So while mww is letting va know that wake word is detected, and while va is preparing itself to listen, some audio info can be lost.
topaz sable
I think that this can be controlled via the "Finished speaking detection" select on your ESPHome assist device.
I think "aggressive" is what you want if you would like to cut down on the delay.
wanton ivy
That controls how long HA/Assist waits for your command to finish (a period of silence), not the delay between the wake word and command
topaz sable
Oh. My mistake then. Does it impact openWakeWord then? I swear someone mentioned it in similar context before.
I figured it would cut down the period after wake word collection that is considered part of the wake word so that it would begin capturing the command sooner.
wanton ivy
I don't think so, but it's been a while since I looked at the code
This is where that's handled: https://github.com/home-assistant/core/blob/dev/homeassistant/components/assist_pipeline/vad.py
slender pumice
Determining precisely when the wake word has finished is very challenging. However, if you disable the wake sound entirely (its just a switch in HA), you do not need to pause much at all.