Sentence Node (beta)

[zachowj]

placeholder

[matczar]

Hi,
Is there a way to make a sentence answer other than "done"?

[zachowj]

Currently, no.

[zachowj]

This was added to version 0.56.0 of node-red-contrib-home-assistant-websocket and requires version 2.2.0 of hass-node-red

[matczar]

Thanks! It was really fast! And I confirm that custom answer works as it should.

[miGii]

Can this custom response use wildcard data as variables?

[donburch888]

Unfortunately the Response is given before the command can be processed ... so asking "What is the weather" gives response of "done" before the flow is started. And when the flow is finished there seems no way to say the result to the satellite device.

[dvejsada]

Would it be possible to have the response as separate node, so it can receive information from the flow? For example, the custom sentence would ask for weather and the response would be formulated using info from weather API?

[zachowj]

I think that would be out of the scope of the node. This is more like a custom conversation agent that would get its answers from Node-RED.

https://developers.home-assistant.io/docs/core/conversation/custom_agent

[danieldioszegi]

I also liked the idea of having separate responses in different use cases. Like

• Turn up the volume
• Turned up the volume / The TV isn't on right now

It would be also great to be able to provide a list of responses and the node would select one to mimic a "conversation" and not replying always with the same phrase.

Btw I haven't opened NodeRed for months and seeing the sentences node in there was such a huge and pleasent surprise.. Great great work!! :) :)

[MrAvana]

Is there a way to implement custom responses that would act like this:

Example config/custom_sentences/en/responses.yaml
language: "en"
responses:
intents:
HassTurnOn:
default: "I have turned on the {{ slots.name }}"

[skylord123]

@zachowj Honestly I find it kinda useless without being able to craft custom responses. 99% of the stuff I would use it for in NR will be to ask a question and craft a response. Even if it is to trigger something if that fails I want to be able to tell the end user.

Really hoping this is changed to facilitate that if possible.

[donburch888]

I agree that Response should not be part of the sentence node, unless you change the response to "let's see", "i'm thinking" or "ummm".

The issue is that there is no way to get a message to the satellite device, in order to give a response.
Wyoming device and Piper seem to be devices with entities - but neither have any services that can be called.

[phellarv]

Is it possible to process variables in the Response-field?
If not, it would be a very nice feature to have.

[kizovinh]

Is it possible to add msg.entity to know which entity the command comes from?

[SpectroHA]

I came here to ask for this so upvoting here.

use-case: I want to handle my own responses down stream as service calls, which is easy enough, but I don't have a way right now to know which entity the sentence is coming from (there are 3 possible sources), so I don't know which entity to call for the response.

[MrAvana]

Which service call handles responses?

[SpectroHA]

just a service call node, domain tts, pointed to a media_player entity.

[salleq]

Is there a way to create sentences like in "intents", so for instance the following:
"Tell me [the] weather [in] [backyard|frontyard|outside]" so that you don't have to type each possible sentence separately?

Otherwise this is already a heck lot easier to implement than in HA by restarting every time when you test something!

[salleq]

Nevermind, I should've tested it first - it works!

[donburch888]

Is there any instruction for getting NR sentence node to be triggered from HA ?
I followed the example, and have the Call Service node successfully triggering HA Voice Assist; but nothing I have tried triggers the NR Sentence node.

[zachowj]

Paste in the details from https://zachowj.github.io/node-red-contrib-home-assistant-websocket/guide/diagnostics.html

[donburch888]

Version: 0.63.1

Home Assistant version: 2024.3.3
Companion version: 0.0.0

Node-RED version: 3.1.7
Docker: yes
Add-on: undefined

Node.js version: v18.19.1 x64 linux
OS: Linux 6.6.20-haos x64

Node.js version: v18.19.1 x64 linux
OS: Linux 6.6.20-haos x64

[donburch888]

Companion version should not be 0.0.0. Should I uninstall and reinstall the companion in HACS ?

[zachowj]

Did you complete step 5 https://github.com/zachowj/hass-node-red?tab=readme-ov-file#hacs

[donburch888]

I guess you are referring to

In the Home Assistant UI go to "Configuration" -> "Integrations" click "+" and search for "Node-RED"

which I think should now read:

From "Settings" in the HA sidebar, select "Devices and Services", click the blue [+ Add integration] button (in bottom right of the page) and search for "node-RED", and install it.

And the answer is that no, it was not installed. I have been using NR with HA for a couple of years ... so not sure when that happened.
The sentence node is now being triggered (though not doing quite what I expected). THANK YOU for your help.

[WhoDunDidIt]

Now that HA added the device_id to the conversation service, it would be amazing if the Sentence node would support it!

[WhoDunDidIt]

Fantastic! I shall test it and let you know!

Thanks so much!

[WhoDunDidIt]

Here's my initial test results:

https://i.imgur.com/fuvbqSq.png

[donburch888]

Node-RED v3.1.8
node-red-contrib-home-assistant-websocket ... ahh updated from 0.63.1 to 0.64.0
HA node-RED addon v 17.0.11 stopped and restarted
Node-RED companion 3.1.6 ... tick

"Hey Jarvis, what is the weather ?"
15 second delay
... and the debug pane shows this slightly before hearing the response "done"

So yes, it passing one of those hidden internal device IDs.

But how to send the actual calculated result "Current weather is Sunny." back to this device ???
And yes, because the response should be sent back after processing, it is not part of the sentence node.

[zachowj]

I am only assuming here but if your satellite device has a HA media player in it you can use the device id as a target of a service call.

[WhoDunDidIt]

I am only assuming here but if your satellite device has a HA media player in it you can use the device id as a target of a service call.

I tested this last night without success. For now I can just keep a lookup table for device id -> entity id. Though I wonder if there is a way to just ask HA to provide the entity id for the given device id.

[donburch888]

Unfortunately knowing the device_id is useless until the device has a way to play a response.

Also is the device_id the hidden HA internal ID; or the user-friendly name of the device ?

[WhoDunDidIt]

Oh I know there's ways to work around it, but I'd rather avoid layering workaround on top of workaround on top of workaround, when the "proper" solution would be to simply pass the device id that is provided by Home Assistant through to the node.

If I had any idea how to set up a development environment for this project I'd take a stab at it myself, it should be fairly trivial.

[donburch888]

My devices CAN play a response though, as long as I know which device I should send it to.

From my research I am guessing that you are sending the TTS to a separate media_player device which just happens to be running on the same physical RasPis (or in the same room) ??
If you have found a way to send TTS to the Wyoming satellite, either through Piper or direct, please tell !

[WhoDunDidIt]

From my research I am guessing that you are sending the TTS to a separate media_player device which just happens to be running on the same physical RasPis (or in the same room) ??

You would be correct on this. However, the Sentence node is able to return a custom response, so it should be possible, I just have no clue how you'd do that.

[donburch888]

Sorry, but the Response field seems to be plain text. It is possible to use different text instead of the default "Done", and I have entered a space to make it say nothing.
And the sentence node happens before I know what response I want to pass back - so the timing is wrong.

[SpectroHA]

My devices CAN play a response though, as long as I know which device I should send it to.

From my research I am guessing that you are sending the TTS to a separate media_player device which just happens to be running on the same physical RasPis (or in the same room) ?? If you have found a way to send TTS to the Wyoming satellite, either through Piper or direct, please tell !

The sauce is something like the below in a function node, which is then sent to a call service node.

A little context on my arrangement. I have 3 atom echos so I have to determine which echo is activated which is done with a state node for each echo, which writes its entity_id to flow.echo. If you only have 1 you won't need this and you can just hard code your media player entity id. After the sentence node and some switch nodes for logic I have a bunch of phrases in a file where a random one is read by an exec node and the phrase is sent to msg.payload, and then handed off to the function node. Then finally handed to a call service node.

var theEcho = flow.get("echo");
var theLanguage;
var theVoice;
var theMessage;
var thePayload;

thePayload = msg.payload.replace(/ /g, '+');
thePayload = thePayload.replace(/[\n\r]/g, '');
theLanguage = "en_US"
theVoice = "en_US-amy-low"
theMessage = "media-source://tts/tts.piper?message=" + thePayload + "&language=" + theLanguage + "&voice=" + theVoice


msg.payload = {
    domain: "media_player", 
    service: "play_media",
    data: {
        entity_id: theEcho,
        media_content_id: theMessage,
        media_content_type: "provider",
        announce: true
    }
}
return msg;