Hey yall, maybe been asked before, but couldn't find anything. Whenever I say "Hey Jarvis" or "Hey Nabu" or "Hey Mycroft," the device will start listening, but immediately cut out and go back to sleep. It's the only real issue I have for it. I've tried the beta firmware and relaxing the listening ending conditions to no success. Anyone know anything else that could help? It's connected to a 3.5mm speaker but that doesn't seem to matter. I also have it connected to chatgbt as the ai.
#Home Assistant Voice Assistant Preview hardware cutting out right after I say "Hey Nabu"
1 messages ยท Page 1 of 1 (latest)
this usually happens if there is something wrong with your speech-to-text setup.
what are you using for STT?
and have you checked the settings of the pipeline that is assigned to the VPE?
Thx for the response, I unplugged the 3.5mm jack to the speaker and it seems to work now. Speech to text is chatgpt. I've looked at the settings to the actual device but dont know much about the pipeline
it not working with something plugged in is odd but the way the pipeline works is that.
STT - converts recording of speech into text
this is then forwarded to
the conversation agent - which processes the text, does actions and generates a responce.
the response is then fed to
TTS - converts the text of responce to speech to playback
Thx for the info at work from home will try to fiddle with it
I turned off prefer local and it still cuts out
Home assistant cloud is the only option I have other than none
prefer local is recomended because if its a basic command line "turn off the light" it means it doesnt have to hit the LLM
OK i turned that back on and it still cuts out, more often if I have a 3.5 speaker connected
I can try another speaker its one of those travel speakers maybe it has a voice component
do you have an active HA cloud sub and a solid internet connection to the HA server?
Yeh I got a sub
And its sitting right next to the router on a 2.4 ghz, it actually runs a lot faster than the reviews
Here lemme make a video
Maybe she just hates me
on settings-voice assistants where the pipelines are. click on the 3 dots next to the pipeline you are using and go to debug
see if there is any errors showing in that
the server should ideally be hardwired btw. wireless for HA server is not recomended
Yeah i have it all on 2.5 ghz
The voice assistant though is stuck to using wifi righht?
Yeah but I can make more
i meant physical device
Oh yeah just 1
is the above screenshots showing one of the "failed" ones. you can see the timestamp at the top
Yeah that's a failed one and the other one is a successful one
you only linked 1 set of screenshots?
Here this is what didnt work
Well hmm
Maybe this all was one command
Sec
This is a failed ond
Is there something other than cloud?
It's the only option I have
you can run whisper and piper locally to handle the speech locally
Are those better?
better is subjective, they are a different options
what type of install do you have?
are you running HAOS?
Yeah
easiest way to set up the local voice addons is to press the "set up voice assistant" button on the VPE device page. this will reopen the wizzard which will allow you to select local and install the addons
speed and accuracy may vary depending on your hardware
OK cool thx I'll check that out
It may not be the actual voice stuff it may be something to do with me using beta firmware and trying to "relaxed" option for finished speaking detection
It's muted right now
well that might be a factor...
It's muted right now not when I was testing
can you comfirm the firmware version? on the device info in the top left
Im just taking pictures of my control tablet lol
Lemme turn off beta firmware one secb
Yeah still turns off prematurely
you can try plugging into a PC and reflashing the firmware using the web tool as described here this will also allow you to see the device logs whilst its attached
Awesome I'll try that, thx
I'm experiencing the same thing since updating a few days ago. Here's the log from my device when it detects my wake word "Ok Nabu"
Basically, it plays the wake sound, then nothing.
I have it adopted in ESPHome, but I have no changes to the config file.
just by adopting it you have made changes to some extent. i recomend you flash back to stock using the web tool. unless you actually need/want to customise then theres no need to adopt into the esphome builder
I have it adopted so I can see logs as I have several other ESPHome devices as well.
Looks like the main difference would be that the binary fimwares are built with esphome 2025.4.0, while I'm building using the latest, 2025.4.2.
the VPE also uses custom versions of some components (until stuff is upstreamed) so total compatibility between versions might vary a little
i have a ton of ESPHome stuff thats adopted but i specifically dont adopt the VPE's because theres no real upside.
although i will probably get another at some point to adopt and mess with customisations etc...
production systems i want to work though ๐
Interesting. I'll try to go back to factory. I did it before because I was diagnosing some other issue with stock firmware.
if you want logs on stock firmware you can still get them by plugging in by usb
what are you using for STT?
I've tried a few different STT services, but no request ever makes it there.
Speaches.ai with a Wyoming proxy as well as Rhasspy are the two that I'm flipping between right now.
and you have the VPE set to pipeline with correctly configured STT? (just ticking off the obvious stuff here)
and do you have any fancy network setup? or just basic
Yup. It's set to use my Primary pipeline. That pipeline has a STT set up.
This part of my network is fancy. Both devices on the IoT network. I'll check to see if any firewall rules changed or are triggered.
it cuting at that specific point is implying it is unable to stream the mic to STT for some reason
yeah it sends to ha which then sends it to your stt provider
Ok. Thanks. The STT server is in a different vLAN which only HASS can access, so that changing could have been an issue.
you could try putting them on the same vlan for testing purposes. i am not 100% sure on exactly how it all gets fitted together but maybe thats causing it to fail
Since I can't test http requests from the VPE, I just put my phone on the vLAN and verified I can curl Home Assistant.
sometimes stuff just doesnt play nice across vlans even though it should
HASS and the VPE are on the same vLAN.
And I know HASS can talk to the STT server because it's running in the same container on the other network as the TTS, which I just tested through the front end.
Well, I donโt know what happened but it seems to be working now. I unplugged the VPE so I could plug in the atom echo I had on my desk. It listened fine. Then I plugged the VPE back in and it works too. I may have flipped a few things back and forth in home assistant or restarted the Wyoming service, but I thought that was all before my last test.