Conversation process variable

Configuration Voice Assistant
1

Hi, I’m trying to accomplish a lovelace button that when pressed, tells the result of an assist (llm) query

If I call this service:

service: conversation.process
data:
  text: dimmi le previsioni meteo per oggi e la temperatura attuale in balcone
  agent_id: conversation.google_generative_ai_conversation

I get this result:

response:
  speech:
    plain:
      speech: >-
        Il meteo attuale a Roma è nuvoloso. La temperatura attuale in balcone è
        29.6°C.
      extra_data: null
  card: {}
  language: it
  response_type: action_done
  data:
    targets: []
    success: []
    failed: []
conversation_id: 01J0TGBVKVH4MRZJZWVHNJFBD1

I want to pass the “speech” part as a variable and tts that variable on a device:

I did this script:

alias: Meteo OP5T
sequence:
  - service: conversation.process
    metadata: {}
    data:
      text: dimmi le previsioni meteo per oggi e la temperatura attuale in balcone
      agent_id: conversation.google_generative_ai_conversation
    response_variable: risposta
  - service: tts.speak
    metadata: {}
    data:
      cache: true
      media_player_entity_id: media_player.oneplus_5t
      message: "{{ risposta }}"
    target:
      entity_id: tts.piper
description: ""
icon: mdi:weather-lightning

Problem is that the message is the full response and not just the “speech” part. How to filter that? I tried {{ speech.risposta }} but did not work

2 
{{ risposta.speech.plain.speech }}
3

it gives this error:

Error rendering data template: UndefinedError: 'dict object' has no attribute 'speech'

the yamls is now:

alias: Meteo OP5T
sequence:
  - service: conversation.process
    metadata: {}
    data:
      text: dimmi le previsioni meteo per oggi e la temperatura attuale in balcone
      agent_id: conversation.google_generative_ai_conversation
    response_variable: risposta
  - service: tts.speak
    metadata: {}
    data:
      cache: true
      media_player_entity_id: media_player.oneplus_5t
      message: "{{ risposta.speech.plain.speech }}"
    target:
      entity_id: tts.piper
description: ""
icon: mdi:weather-lightning
4

Sorry, I missed the top-level of the result. It’s probably

{{ risposta.response.speech.plain.speech }}

You can look in the trace for the automation and see the “changed variables” tab for the step that calls the service to see the actual structure of the data.

5

thank you it now works!!