#Home Assistant Voice, ring lights not working.

1 messages · Page 1 of 1 (latest)

zealous bluff
#

I bought the HA Voice a few months ago and I am STILL trying to get it to work. I tried tackling it again tonight. When I said the wake word "hey jarvis" I hear the acknowledge tone, but the lights on the top didn't light up. I unplugged the device and plugged it back in and I got the initial white lights, followed by the flickering blue lights (indicating it was connecting to wifi). Again, I said the wake word, and then told it to turn the lights off... the ring light did not illuminate, and i never got a response.

I loaded up the integration and I can see that it recognizes the wake word (confirmed with both the status change within HA as well as the wake sound) but after that, nothing....

zealous bluff
#

I am still having this issue. I hope my device isn't defective.

twin horizon
zealous bluff
#

Actually, I've realized that my Whisper isn't connecting to the Wyoming Protocol. I have tried to get it connected but it's not working. I even Uninstalled and reinstalled Whisper. Piper is connected to Wyoming, but even with Whisper running, I can't get it to connect. I'm not sure if this is what's causing my other issues, but may explain why I haven't been able to get the assistant to work yet.

twin horizon
zealous bluff
#

I have Whisper (and Piper) installed through the Add-ons and I have Wyoming Protocol installed through the Integration.

twin horizon
zealous bluff
#

My HA setup is run on a Raspberry Pi4. Whisper, for some reason, takes like 10 mins to start. My HA just restarted, so I was waiting to see what the logs say. I don't see any errors, I do see a lot of "exit code 9" but the text is in Green. My Whisper is currently running, but when I try to add it via the local reserved IP I have for my HA server, or even trying "localhost" and using port 10300, I get Failed To Connect

twin horizon
#

you could run speech-to-phrase instead which would be a bit more limited functionality but would have a better chance of working.

#

another option is to use a STT from somewhere else. you could run whisper on another machine for example. ideally one with GPU acceleration as an option.
or you could use a cloud service. e.g. home assistant cloud (from Nabu Casa) . but this means its not fully local and has a price.
HA cloud comes with some other benefits and also directly supports the project. but it depends what your requirements are.

zealous bluff
#

Ya, I was trying to set it up to use ChatGPT, but that costs too, and also means it's not local (which is one of the many reasons I'm trying to get away from Amazon. I don't have any hardware that can run an LLM, so I may just end up going with Nabu Casa. At least that will be supporting HA.

twin horizon
zealous bluff
#

So if I go with Nabu Casa, I won't need Whisper and Piper installed in my HA instance?

zealous bluff
twin horizon
twin horizon
# zealous bluff Ah....hmmm

the pipeline is
speech - STT - text - brain - response - TTS - output speech
whisper and piper are for STT and TTS and they can be replaced by NC cloud but the brain will still be the basic one unless you add something else in there.

zealous bluff
#

So without Whisper (or another STT) would that explain why I don't get the blue "listening" lights after I say the wake word and hear the wake sound?

twin horizon
#

it knows its being asked to do something but it cant do it.

#

for some more detailed stuff on how the voice pipeline works you can check out this which was streamed a few weeks ago - https://www.youtube.com/watch?v=sIkguv0NEQI
perhaps more information than what you need but if interested then its pretty cool.

#

if your only needing basic commands then the built in "brain" is okay but it is a little bit specific about how you word things.
you should try out the speech-to-phrase addon instead of whisper and set that up as your pipeline. for basic commands it works pretty well even on something like the RPI.

#

STT (whisper) tries to provide an exact transcription, whereas speech-to-phrase generates a list of phrases based on your install and then matches the incoming speech to the best fit as far as it can tell.

zealous bluff
#

Ok thank you. At least I have an idea as to what's going on and at least I know my device isn't defective 😀