Capture assist/piper output text (and then speak it on chosen output)

Configuration
,
1

So I picked up the HA Voice Preview Edition, and it works well enough to capture and process commands, but the speaker is too anemic to be heard unless I’m up close to it. Can’t we send that output elsewhere?? I can trigger an automation off of tts.piper changing state (it seems to simply change to the last time it did something) or assist_satellite.home_assistant_voice_09e66a_assist_satellite changing to “Responding”, but I can’t then find an attribute of any of these entities that shows the output text (short of setting everything to debug and then digging through the logs). The logs refer to a chat_log but this doesn’t seem to be accessible anywhere.

I tried things like this, which don’t work:

- id: 'id_19.1'
  alias: "Capture TTS Text"
  trigger:
    - platform: event
      event_type: call_service
      event_data:
        domain: tts
        service: speak
  condition: []
  action:
    - service: input_text.set_value
      target:
        entity_id: input_text.say_this
      data:
        value: "{{ trigger.event.data.service_data.message }}"

Kind of irrelevant to the the first part, but the rest of the setup is this: I’m running in UTM on a Mac Mini, so I don’t need the cloud for local processing, and this Grok integration handles more complex queries really well. I have an automation that watches input_text.say_this; when it changes, the automation ssh’s a command to the host OS that uses the MacOS stt commandsay to speak the text to whichever AirPlay speakers are attached to Airfoil at the time.

Maybe this is an enhancement request or is already in the pipeline?

2

I’ve found that when you ask this question everything goes quiet… :grin:

If you use custom sentences and intents you can send the output anywhere you like, but for built-in stuff I think the answer is no.

3

Seems right. I’ll go file a feature request

1 Like
4

There are multiple ways to do it. In developer perspective, you can simply save the response in a state variables. You can mod any of integrations => wyoming protocol (example below), openai integration or piper addon etc…

self.hass.states.async_set("wyoming.response_text", event.data["tts_input"])
or
self.hass.states.set("hello_state.world", "Paulus")

Yeah, it is your lovely first integration on the HA tutorial

e.g. Injecting code in Wyoming protocol assist_satellite.py:

image
image744×299 36.9 KB

STT: How are you today?
Output:

image
image646×168 14.9 KB

One last thing

Don’t forget to create an automation with above state trigger. :smiley:

alias: Response Text Automation
description: ""
triggers:
  - trigger: state
    entity_id:
      - wyoming.response_text
    from: null
    to: null
conditions: []
actions:
  - action: notify.persistent_notification
    metadata: {}
    data:
      title: TTS Response
      message: "The response is: {{ states('wyoming.response_text') }}"
mode: single