Open Greylinux opened 7 months ago
After watching the latest HA year of the voice 6 I had an idea related to this issue, my suspicion looking at the log is that its the delay after the wake word is spoken to streaming commands to HA. If I pause longer than I do with the directly connected speaker it doesn't cut off the first word and therefore the accuracy of the satellite is as good as the direct speaker. Is there a way to change the delay after the wake word is spoken using the installer?
I have the same experience, it would be useful to reduce the delay.
so I'm having an issue with the accuracy of the wyoming satellite.
The setup I have is, a anker powerconf S330 connected directly to my Beelink Se12 running HAOS, Rhasspy repo version of Assist microphone addon and whisper, piper, openwakeword running as addons also.
The satellite is a Raspberry pi 4 also with a anker powerconf S330 connected, I used the installer to setup the satellite with the recommended settings from the tutorial.
The conference speaker connected directly to my HAOS server has near 99% accuracy and is super fast which is great, the satellite however has closer to 50% accuracy.
this is the whisper log for the connected conference speaker :
this is the log for the wyoming satellite
The settings I have for the assist microphone addon are :
VAD : silero Noise suppression: 2 Auto-gain: 15 Volume multiplier: 1
The settings for the satellite using the installer
VAD: ? Noise suppression: medium Auto-gain:15 Volume Multiplier: 0.1 ( in HA device settings) 1.0 in (satellite installer settings) not sure whats going on here or if its even the same thing as its called MIc volume in HA Device settings. I did try changing it to 10 ( the max) in HA device settings with no advantage.
Any idea why the big difference in accuracy?