[REQUEST] Support choosing response media_player in config

[Ramblurr]

Note: this is not a duplicate of issue 316. Issue 316 is about using the AUX/3.5mm audio out on the PE for music and the builtin speaker for Assist's response.

This request is about configuring the PE such that the TTS response Assist generates can be configured to go to a different media_player entity entirely.

Apparently this is already possible for those building custom esphome voice satellites by specifing voice_assistant.on_tts_end in the esphome config.

• https://community.home-assistant.io/t/year-of-the-voice-chapter-5/655319/25
• https://community.home-assistant.io/t/voice-assistant-have-assistant-reply-on-a-different-media-player/625875/6

[BlackandBlue1908]

+1

[kosmaluk]

Popieram. Funkcja wyboru media playera do ogłaszania odpowiedzi jest bardzo pożądana

[kosmaluk]

you can easily take over the code and make changes without losing the ability to update. A new Home Assistant Voice Firmware update sensor has been added. You can add your own media_player or create start/stop wakeword buttons and in automations make a continuation of the conversation, dialogflow, etc.

[Ramblurr]

you can easily take over the code and make changes without losing the ability to update

Can you describe how to do this? It isn't obvious to me. The PE doesn't show up in esphome builder.

[kosmaluk]

You need to enable TAKE CONTROL and install the device in Esp Builder. If you don't see it, enable the three dots at the top and "show discovered devices". After installing, you can add any elements you are interested in to the section. For example, mediaplayer or other commands that you will use in automations

on_tts_end: - homeassistant.service: service: media_player.play_media data: entity_id: media_player.google_home_mini # <- change this media_content_id: !lambda 'return x;' media_content_type: music announce: "true"

[phila-dot]

Have you successfully done this or just assuming because it works with the older VA’s?

I keep seeing people post this, but it doesn’t work with the VA PE nor the new build S3 Box. There is nothing returned, so you get no tts audio.

[kosmaluk]

If I write a solution, it means I did it and it works. Why wouldn't it work? The documentation for ESPHome is the same and applies to all these devices

https://esphome.io/components/voice_assistant.html

You have to Adopt a device and adapt the code. The examples that are on the forums work perfectly.

I have collected various solutions and together with ChatGpt I have implemented the solutions that I wanted. I have done, among others:

1. Sending sound TTS to another media player
2. Question and answer sensors (I use them to debug STT operation and automation to continue the conversation without waking up again)
3. START and STOP wake word buttons, which I also use for these automations

Everything works great.

Everyone got scared by the entry in the documentation not to take control and are waiting for new functions.

In the movie Chapter 9 they said that they added a new Firmware update sensor which you would normally use to update the device after taking over the code

[Ramblurr]

You need to enable TAKE CONTROL

That's the bit I can't find. I have seen other people mention that. But I can't find that option anywhere. Where in HA is a "take control" button?

[kosmaluk]

something is wrong with your eyesight :).
I wrote how to do it a few posts above :)
ESP Builder --- enable the three dots at the top and "show discovered devices". HAVPE will show up and click ADOPT

[Ramblurr]

something is wrong with your eyesight :).
I wrote how to do it a few posts above :)

I saw what you wrote, but that post don't contain the string "take control", so I assumed that entire post was for after you "took control".

I appreciate your assistance here by the way.

ESP Builder --- enable the three dots at the top and "show discovered devices". HAVPE will show up and click ADOPT

Ok, now we are getting somewhere, the actual button is "show discovered devices" and then "adopt", not "take control". I appreciate the exact instructions.

However, when I click "Show discovered devices", nothing happens (except now the button says "Hide discovered devices". The list of esp devices remains the same, and its just my own custom devices.

Presumably this is because my IOT devices and HA are on different vlans. Is there a way to import it manually without relying on MDNS?

Edit: I managed to do this myself, I simple created the yaml config I wanted, and ensured the wifi.manual_ip component config was set to the static ip of my existing Voice PE device, then installed it using esphome builder. Pretty straightforward. I copied https://github.com/esphome/home-assistant-voice-pe/blob/ba4fca5344724d56b586f7ea9173e0cb2db2773f/home-assistant-voice.factory.yaml as the base yaml, and then added my wifi overrides.

I still think that is feature request stands however, as this should be possible to configure without having to edit esphome firmware.

[itispip]

Everyone got scared by the entry in the documentation not to take control and are waiting for new functions.

Exaclty, that's my fear. Even "taking control" is easy, and ESPHome added a new Firmware update sensor, but what I'm afraid is update new official firmware will need back to factory firmware first, then update, then ESP take control again, and then customize/add the sections not available in factory firmware again --- that can be a repeated pain.

[kosmaluk]

I don't know, but I updated the device to 2025.2 and it went without a problem and my configuration remained. We'll see how they implement some new functions or elements or add them

[ev1502]

If I write a solution, it means I did it and it works. Why wouldn't it work? The documentation for ESPHome is the same and applies to all these devices

https://esphome.io/components/voice_assistant.html

You have to Adopt a device and adapt the code. The examples that are on the forums work perfectly.

I have collected various solutions and together with ChatGpt I have implemented the solutions that I wanted. I have done, among others:

1. Sending sound TTS to another media player
2. Question and answer sensors (I use them to debug STT operation and automation to continue the conversation without waking up again)
3. START and STOP wake word buttons, which I also use for these automations

Everything works great.

Everyone got scared by the entry in the documentation not to take control and are waiting for new functions.

In the movie Chapter 9 they said that they added a new Firmware update sensor which you would normally use to update the device after taking over the code

Thank you for posting screenshots, are you by chance able to upload to upload your full configuration file so we can see all the details of the progress you've made?

[Ramblurr]

Here is my esphome config for my Voice PE. Note I've changed from calling media_player.play_media for the TTS audio to doing instead assist_satellite.announce.

This is because I have one Voice PE that I want to act as only a microphone (that's the one the config below is for). Then I have a linux computer running a wyoming-satellite which is plugged into my whole home audio streaming system with a hardware ducker.

This lets me speak into the Voice PE, but hear the TTS replies over my built-in wall speakers.

You have to be running 2025.7.3 at least

...normal api/wifi/logging/etc that is in every esphome device is omitted...

packages:
  Nabu Casa.Home Assistant Voice PE: github://esphome/home-assistant-voice-pe/home-assistant-voice.yaml@dev

voice_assistant:
  id: va
  media_player: !remove
  on_tts_start:
    then:
      - logger.log: "on_tts_start-start"
      - homeassistant.action:
          action: assist_satellite.announce
          data:
            entity_id: assist_satellite.my_satellite
            message: !lambda 'return x;'
      - logger.log: "on_tts_start-end"
  # or, if you want to use media_player.play_media...
  # on_tts_end:
  #   - logger.log: "on_tts_end-start"
  #   - homeassistant.action:
  #       action: media_player.play_media
  #       data:
  #         entity_id: media_player.your_player
  #         media_content_id: !lambda 'return x;'
  #         media_content_type: music
  #         announce: "false"

[lextiz]

Thanks for opening this. This is exactly what I am trying to figure out before building a PoC.

My setup would be Voice PE as mic / wake-word only, with replies played through a normal HA media_player (Sonos first, but ideally generic). I am also hoping continuous conversation can still work, so HA would need some reasonable way to know when the external speaker has finished before listening again.

The on_tts_end YAML workaround is useful, but I would rather not take over the firmware if the right fix belongs in HA / assist_satellite.

Do maintainers see this as a Voice PE firmware option, or should this be proposed in HA Core as an assist_satellite response output target?