LLM Script for Music Assitant voice requests

Configuration Blueprints
1

I’m using the Blueprint “LLM Script for Music Assitant voice requests” from @TheFes .
It works most of the time but I’m experiencing 2 issues right now.

  1. When asking LLM to play something it stops the current playback and finds the requested title but does not play it although the LLM responds that it does.
  2. Sometimes when asking the LLM to play something it responds that is does but actually nothing is played and MA did also not change to the new title.
2

Hi!

Which LLM are you using, is it a cloud LLM, or a local one?

To see what goes wrong, I would have to see the traces from the script runs. Can you share them by uploading the json of the trace to a code sharing site like https://dpaste.org.

You can download a trace using the download button in the upper right corner.

I would need to see the traces of a run in which the result was not what you expected, so when it didn’t start playback.

3

Hi, I’m using the Google AI cloud which is integrated into HA core. I’ve uploaded the trace. This is from when HA selected the correct title but did not play it.
https://dpaste.org/OnZ3F

I asked LLM to play “something from Pink Floyd”. The the same result is when asking “play Pink Floyd”

4

The response from Gemini looks fine here

            "media_id": "Pink Floyd",
            "artist": "Pink Floyd",
            "media_description": "etwas von Pink Floyd",
            "media_type": "artist",
            "context": {
              "id": "01JJP7FQSHTWEA3JQCKE7SVVTT",
              "parent_id": null,
              "user_id": null
            },

Action data also looks fine

              "domain": "music_assistant",
              "service": "play_media",
              "service_data": {
                "media_id": "Pink Floyd",
                "media_type": "artist",
                "artist": "Pink Floyd",
                "radio_mode": false,
                "entity_id": [
                  "media_player.receiver"
                ]

What happens if you do this in developer tools > actions

action: music_assistant.play_media
data: 
  media_id: Pink Floyd
  media_type: artist
  artist: Pink Floyd
  radio_mode: false
target: 
  entity_id:
    - media_player.receiver
5

When I call this once it works. When I call it again id does not. When I call it a 3rd time it works again and so on. Strange. Any idea?

6

You are referring to using the action in Developer tools > Actions now?

If that is the case, the problem is in Music Assistant. The script simply performs this action. This is something you need to address to the MA guys then, that part is a black box for me as well :slight_smile:

7

Yes developer tools → action. I will create a MA bug report. Thanks for your support.

8

Actually I figured it out. It has to with how the MA player queue is managed. If the enqueue options are set to play then the issue goes away. These can be configured in the core system settings.

1 Like
9

Hello @Starob. I have the same error 1 that you describe. Where can I change the MA Player queue in the core system. Unfortunately I can’t find it.

10

@TheFes would it be possible to update the script to identify a specific label and use that label as the targeted media player in a given room?

Then user just ensures that only one media player in a given area has that tag and then filter on room+tag? If area is targeted no more confused llm. It’s kind of like your default player idea but more dynamic…

Man that would make things so easy…

11

Look here:

mp
mp2179×1101 155 KB