#Announce is acting up!

1 messages · Page 1 of 1 (latest)

river abyss
#

I have a Voice PE in the Living Room and a few Atom Echos arround the house.

If for instance in the office i say "Ok Nabu tell me a joke" the HA always annouces on the Living Room PE instead of replying on the device that i currently talked to and i dont seem to be able to find where to change that.

Also unrelated question: if my Ollama instance is not running and i ask a question to a Voice pipeline that as AI he should just try to use local Assist right? because mine starts the ollama instance on another machine by itself 😄

#

I tried with the same "Voice Assistant" and with 2 separate pipelines

#

Oh and devices are not in listening range from eachother 😄

broken anvil
#

some models do this stange thing where they talk to different devices. try changing model back and forth

river abyss
#

since i have powerfull AI running i just asked Nabu why are you replying to me in devices all over the house

#

and it replied: " From now on ill just reply in the device i was asked" 😬

broken anvil
#

sometimes it seems because its a "voice assistant" it thinks it should speak to everyone

#

messing with the prompt has had varying success

river abyss
#

i was using the 3.1 8b

#

ill switch up to 3.2

#

Here do i need this?

#

one for each device?

broken anvil
#

you shouldnt need to have seperate pipelines no unless you are doing different things with them

#

i have different pipelines connected to different ollama connections with different prompts for different personalities. but for general usage with the same prompt you should be able tro share it

teal sundial
#

You can get the device_id from the trigger and use that if for the response device

#

trigger.device_id has the id of the device that caught the sentence. Use that to create your responses on that same device

river abyss
broken anvil
river abyss
#

If i talk to the Atom Echos, they annouce in the PE and then also talk, if i talk to the PE he does whatever i ask but doesnt even reply

broken anvil
river abyss
#

yeah because its so random 🤷‍♂️

#

May be getting somwhere

#

If i ask on the atom it annouces in the living room PE, if i ask on the living room it says that it doesnt have media players to annouce and just dictates there

broken anvil
#

can try adding to the prompt to not call any tts tools and just respond in plain text

#

then the plain text will get tts'ed by the pipeline

river abyss
broken anvil
#

no worries, its a bit of a random one that and different models work in different ways and change their mind 1 day to the next

real swift
#

I had constant issues with LLama3x broadcasting to all PEs or random PEs. I switched to Qwen2.5 and haven't had a problem since.

river abyss
river abyss
broken anvil
river abyss
real swift