I’m aware the Raspberry 5 is not yet certified for running Home Assistant. But might it be acceptable for running the voice assistant satellite code?
I’m currently running the satellite on an RPi 4 + Anker S330 speaker and it works…okay. It can take a while to detect speaking has ended, or sometimes fails to acknowledge it at all.
I’ve got the homeassistant-satellite running (I’ll be re-trying the wyoming-satellite this week) along with local wakeword. Wakeword is configured with a threshold of 0.8 and trigger level 3. I’d like to increase those numbers in hopes of reducing the too-many false triggers, but suspect more processing power might be necessary than the RPi 4 offers.
If you run satellite/wakeword on an RPi 5, how well does it work for you?
I use RPi 4 with Anker S330 and it runs fine with wyoming-satellite. I used homeassistant-satellite before and it was by far not as good as wyoming-satellite.
I’ve now migrated to wyoming-satellite, but can’t say I notice any appreciable difference in behaviour.
starob, do you run wyoming-openwakeword locally on your RPi 4? I had found with homeassistant-satellite that local wakeword improved performance, and just continued with wakeword local after the move to wyoming-satellite.
It appears that “it can take a while to detect speaking has ended, or sometimes fails to acknowledge it at all” (mentioned above) might be more a factor of wyoming-faster-whisper STT, which I run independently outside of Home Assistant on my HA host.
I super enjoy the implementation of voice in HA, but it doesn’t work quite predictably enough yet where all adults in our household want to use it. I’m hoping some hardware upgrades will close the gap.
I am having this same issues wit this same problem, I have a RP5 running the Wyoming satellite driver and running into issues getting it to capture sound. Do you know if the RP5 is compatible with this project?
I did have an opportunity to upgrade to an RPi 5 this past Spring. I’ve had no technical problems with it for this purpose.
As a voice speaker the 5 works reasonably well, though seemingly not better than the RPi 4, and perhaps slightly worse. It’s subjective, but the 5 seems to wait even longer for speech to end before forwarding it to the STT process.
What are the symptoms you’re experiencing – is it wake word detection or request capture (or both) that are giving you a problem? Have you tried adjusting the respective process parameters?
Hi Thanks for the reply. I actually switched gears, and used the Pi 5 to install the home-assistant server. And by installation was a breeze. However, since I had a Pi zero 2 w, I installed the wyoming-satelllite driver and everything went well as far as the installation, but when I ran to capture a test.wav file, and replayed the wav I get back this soft white-noise sound from the attached speaker.
Have you verified you’re providing the correct sound device to wyoming-satellite (using the arecord & aplay -D option)?
Or is the linked procedure how you produced test.wav?
Thanks for the reply. I actually got it going to run a test recording. But after a little tinkering, I figured out that I had to be in the correct directory, prior before running these arecord + aplay commands. Before; I was just in my home/user/wyoming-satellite:$ But when I cd into the this line directory; home/user/wyoming-satellite/sound:$ and ran those previous commands I was able to record a test sound and play it as well loud and clear. I am not sure why in the Wyoming instructions they don’t mention this or seems that the instructions are incomplete as it seems. I am following no bringing the satellite service into Home Assistant server describe here: Link The satellite shows up and added, but when I run it with my wake word “hey_jarvis” the sensor does not indicate that its listening. So gotta do a little more tinkering.