When testing changes to the voice assistant, it would be really helpful to be able to remove all of the previous messages in the context. Sometimes I need to test on an actual device because an automation depends on the area the voice command is triggered.
The issue I’m having is that sometimes there are lots of old messages in the history, which affect how the assistant will respond.
Is there a way to get rid of them?
I’m using the Ollama integration for the natural language processing.
Press the “restart” button. You will find it on the device page. This is a fairly quick action that resets the conversation ID.
If you use a chat window in the dashboard, the conversation is reset each time you close it.
I don’t see a restart button.
It’s on the config page for the device…
And the context times out for a voice station after 5 minutes.
Ahh you physically restart the voice assistant?
Thank you, I didn’t know it was a 5 minute time out either.
Yes that’s why I use my phone handset for voice debug.
I can close the chat window new context. While on the VPE you must restart or wait for 5min.
@mchk (and why when voice gets horked on the companion it’s a big deal - see my open voice bug on companion, I’m going to switch to one of your Wyoming integrations to try to see if I can find it… I’m convinced it’s on the phone side 2026.2.x didn’t fix it)
I do the same thing, but sometimes I need to test it with the actual device, if it’s going to do things like play something back on the device that received the command.
I have to say, all of the debugging tools are really good and there are a lot of very useful things for testing this stuff. I’m really impressed with what they’ve done!
Please tell me more about when the problem occurs. I tested streaming in the January version of the app and didn’t find any errors (I use Piper).
As far as I remember, you’re using Kokoro. Which version exactly? The most popular repository hasn’t been updated since spring and doesn’t work with the chunk stream (perhaps this is the cause of the problem).
I recently experimented with uv and built the modern version of the Kokoro server—check its stability using it. I also like the supertoniс project; it’s very responsive, although the voice quality isn’t as good.
I’ll ping you and send the config to keep this on topic here. I’ve got about 6 different ones running right now i have a non streaming kokoro as primary. But it doesn’t matter.
If I use a Wyoming TTS it just doesn’t play audio. If I use OAI compatible (cloud, local doesn’t matter, I’ve got three integrations installed- they a act the same) it just hangs as the text gets sent to the TTS.
All of it just works in browser (three tested) or on a VPE. It’s only the handset.
