Option 3: playback orders?

[guillow88]

Hi, have option 3 with Gemma3 27b running.

Interestingly, it works to play music, but it reports back that it has failed.
However, I'm not able to STOP playback. I've tried "stop playback" "stop music" "stop music in the office" etc. to no avail.

I am able to "Skip to next song" > Response - "Playing next" - plays the next song
"go back a song" > response "going back" - does nothing
"play previous song" > LLM: "playing previous" - goes back to previous song
"stop music" > LLM: "stopping music" - nothing
"pause music" > LLM" Pausing music" - nothing

Other than an LLM intelligence problem...

• is there anything else I should make sure is exposed or assigned in MA or HA?
• are there phrases that others have gotten to work more consistently?

[guillow88]

Wanted to provide an update.

I tried openAi GPT-4o mini > next, previous ok. Unable to stop/pause
tried the more expensive GPT 4.1 > next, previous, pause works! (when I say to stop, it says it cannot)

Perhaps there is more that can be done to more clearly expose the stop function in the script to help the LLM's attention?

[nikito]

Media control commands are not part of these blueprints, those are actually native voice command built into Home Assistant. There should be play, pause, next/previous, and volume control. There's no stop command in HA presently.
I also use gpt-4o-mini and haven't seen any problems with these commands, though i also have "prefer handling commands locally" enabled on my pipeline, maybe that is why.

[guillow88]

Thanks, I have prefer local as well. I'm using a offloaded whisper-large, and kokoro for speech processing. Perhaps I'll add it to the base prompt to help it's attention.

Is there anyway to see a log of what the LLM is actually doing during a recent pipeline? I'd like to see what commands it actually passes and processes. U haven't been able to find anything like that in logs... Perhaps I'm not looking in the right places?

[nikito]

You'd either have to have ollama in debug mode and check logs there, or ask the llm what tools it tried using and what arguments it used. Can also check the script trace log to see what it ended up with. 🙂

[guillow88]

interesting, i think I know why my gemma3 is giving an error even when everything is working, it seems to be related to a script error... Also, gemma3 is choosing to understand pause today, so will go back to local and see how that works...

I've had ollama debug running for a while, and have never found the logs specific to the LLM outputs.

Settings > System > Logs

Top right selector only has HACore, supervisor, host, dns, audio, multicast, zwave, matter server, whisper, piper, file editor, get hacs, music assistant server...

Nothing from Ollama...

I can see attempted actions and errors within music assistant server.

How do I see what Ollama or openAI is doing?"

[nikito]

So i meant within ollama itself (like the docker logs for instance if you are running it in docker), HA doesn't log what ollama does unfortunately, same for open ai. In lieu of that, asking the llm "what took did you use and what arguments did you provide it?" Can sometimes give some insight. The script trace log also she's what the arguments ultimately became from the llm as well.

[Erotemic]

Not having a way to stop the music is a big issue.

There's no stop command in HA presently.

@nikito Perhaps this is new, but using websockets I was able to manually call the media_stop service on the media_player domain using a target with {"entity_id": <my_ha_voice_entity_id>}, so that ability certainly exists.

Now I just need to figure out how to connect it up to the voice commands.