I just received my Voice PE today and I am stumped as to how I can control my devices with it. I show Assist is exposed to every single device I have yet when I tell Nabu to turn on the lights, it says: “sorry, I am not aware of any device named XXX”
Is it just me, or did anyone else find it a tiny bit difficult in setting this thing up? I first tried using my desktop and adding it in Devices, but it just spun around and around and around. After 3 attempts, I set it up using the companion app on my phone and was able to proceed.
The box didn’t even come with any instructions or “quick setup guide”. Just a small card with a QR code/URL that takes you online for the instructions. I think they could have a small quick start guide included - that would be helpful for many.
Anyway, about my devices, what am I not doing, in order for Assist to see them?
The more devices exposed the less accurate it will probably be.
The base set of commands work ok but they must be properly recognized and that is very difficult task. If you had alexa 10 years ago you would understand how long it took to get it to work.
look at the debug for you voice assistant in settings. Verify it properly heard your command.
if it heard your command correctly make sure you have the assistant set to an Area and a light in the same Area. If you dont it wont know what light to turn on.
The Voice assistants for HA are improving but still need patience. if your using HA cloud for STT that improves things as it cleans up bad command it seams but even then it is not perfect.
I have my Voice Preview Edition since today and
I’m seeing the same behavior.
Text pipeline works perfectly for the same commands, but voice commands fail when:
Friendly names are long/complex (e.g. two bedside lamps in the same room require differentiation), or
Multiple lights exist with the same friendly name in different rooms (same type of light, e.g. ceiling light in several rooms.)
Example:
Entity: light.andies_nachttischleuchte
Friendly name: “Andies Nachttischleuchte”
Room: Bedroom
Text pipeline:
“Andies Nachttischleuchte in Bedroom turn on” → works
Voice:
Same command → “does not exist in Bedroom”
This strongly suggests the issue is between speech-to-text and entity/area matching, not in the intent handling itself.
Workarounds to be tested next:
Short aliases
Template lights
But ideally, voice matching should behave as reliably as the text pipeline without requiring aliases or renaming.
Seeing that a device sold for over 70 Euro is so unstable / bugy, although on the market for approx. a year and seeing that this issue was addressed months ago is kind of disappointing.
I hope nabu casa developers are watching this channel.
Voice matching an out of context guess. It’s not a human determining what was said and utilizing context clues, it is spicy autocorrect trying to match words to predefined sentences.
If commands work when typed into assist but fail with voice command you must check STT to determine if it “hears you correctly”. Check the logs of STT provider or you can look at HAVPE logs in esphome builder to see STT output.
Today I had someone trying to determine why “turn on the TV” command did not work. Checking logs they found STT was outputting “turn on the T V”.
Verify STT output is what you expect.
You may need different STT provider or consider more relaiable terms for commands.
problem is not device, it is software.
Software is free, not an excuse, just think it wise to understand limitations and difficulties of voice assistant and regulate expectations. Amazon first 3yrs were terrible. It’s better now but is also back by a data center. Your trying to recreate same experience on a small PC. It better than bad but still not good and maybe 1-2yr from no complaint usage.
I will add, wake word is absolutely amazing. I couldn’t beg Alexa to wake up when Amazon first released echos but HAVPE or any esp voice device with micro wake word is pretty consistent. Voice command is good at short range but poor after 5’. This is limit from my experience.