#How to disable to automatic "Unable to get response" response?

1 messages · Page 1 of 1 (latest)

hybrid wagon
#

I'm attempting to force my assistant to not reply at all so the speakers are free for a script to use them. But when I set the conversation response to nothing by writing it in the prompt, the assistant responds with the "Unable to get response" message.

How can I disable it?
I'm new to all of this, so please don't hesitate to ask for missing information.

hybrid wagon
#

I still didn't find a way to make this work.

In my prompt I included "On action_done, activate the script named 'script name'.". (the script itself plays a positive sound to indicate the action was successful) But the agent follows with a reply, which I want to get rid of. By adding "Respond with 'word of your choice'", I can set the reply to something specific. But if I force it to say nothing, by changing the word for a space, then the response becomes "Unable to get response.". The same happens when I tell it to reply with an empty string or with similar techniques.

Note: the script also includes a "set response of conversation to:" set to reply with a space. But the agent still attempts to respond, sometimes even to say that it successfully ran the script.

Anyone has a way to turn it off?

astral reef
#

@hybrid wagon if you didn't found an answer, try "*". A single Asterisk instead of space

hybrid wagon
astral reef
#

@hybrid wagon I put this in my prompt and its working flawlessly:
You are a voice assistant for Home Assistant.
Answer questions about the world truthfully.
Answer in plain text. Keep it simple and to the point.

I will give you either commands or ask questions. If what you hear is not a question or a command it may be background audio from a tv or another person. If it is not a command or question you should just ignore the request and respond with a string that contains only an asterix like "*". This is will be a voice response so don't include any hyperlinks or references that would not provide value if read.

hybrid wagon
#

Also, that is an interesting method to remove background noise you got there. Would it not give you trouble with questions it doesn't interpret as questions for reasons such as the STT did not add a question mark?

astral reef
#

Maybe is TTS issue, I would look at documentation of chosen TTS

round radish
#

Yeah, * is definitely the right character, I have something similar and it works well:

Decision Hierarchy (process in order):

  1. Questions — Any input with question marks, interrogative words, or seeking information is a question. Answer it. Exception: incoherent nonsense with a stray question word is not valid.
  2. Clear commands — Execute device commands and report final state.
  3. Generic commands — If the area is known, check how many devices of that type exist in the area. If only one, execute and report final state.
  4. Ambiguous device commands — The input is clearly a device command but the target is missing. Do NOT call any tool. Ask "Which area?" or "Which [specific device type]?" (2-3 words max).
  5. Everything else — Input that is not a question and not a device command (nonsense, fragments, not directed at you, unrelated topics): respond "*". If it seems like a genuine attempt to communicate but is garbled or incoherent, respond "Can you repeat that?" instead.

Clarification rules:

  • One clarification only. Nonsense follow-up gets "*".
  • Respond "Okay." for nevermind/stop.
hybrid wagon
#

Thank you both for your input @astral reef and @round radish, I will look further into this. So far it's the best path I have.

hybrid wagon
#

`Hi @astral reef and @round radish. A few questions to help me investigate the weird behavior of my agent when replying with a *.

  • Do you use Local LLM and if not what do you use instead? This might be affecting it.
  • I assume you both use Piper as TTS? (I do)
  • Where did you learn * was the default character to stop it from replying?`
#

Code formatting due to the Asteriks which were messing up my questions.

round radish
#

I use local LLM (Gemma4), I use KokoroTTS for TTS. I learned * after trying " " which often casused a static sound with TTS, and other charecters were confusing or difficult for the LLM to output.

hybrid wagon
astral reef
#

@hybrid wagon I use Gemini or Claude via API and Microsoft TTS and STT, no local TTS and STT works well with Serbian language

hybrid wagon
astral reef
#

I started to assume that * is default when HA companion app got wake word and started to reply only with * if no response was set in given command that LLM performed, so command was not going through any automation or HA built in intent

#

That * is only visible on HACA as a written reply, all other VA devices that do not have screens just do a command and do not say anything

hybrid wagon
#

Change of plans, since I'm still stuck on this and I didn't find a way to make symbols work in my system:

Is there a tool that simply ends the conversation that exist? Google's AI says that conversation_stop is a thing but it's false. But something similar would be ideal.

astral reef
#

Really can't think of anything that can help you, sorry