Is it possible to "tell" voice assistant to instruct something of Alexa?

[opal zephyr]

So, in my journey of constant disappointment relating to Echo devices, I've kinda settled on having a "m5stack echo" device as the input and my amazon echo device being the output of all questions asked of it in my voice pipeline.
Trouble is I still have the issue of music. I suppose i could just "get by" with having 2 different wakewords (hi jarvis for any AI/HA related stuff and Alexa for anything music related) but it seems a bit of a faff. So I was wondering if there was any ability as it stands to be able to send instructions to an echo device , ie "hi jarvis, ask alexa to play music by Queen on spotify" so effectively the Alexa device would be happy to play music since its "asking itself" if you know what i mean.

long shot, i know, but i don't know if there was any kind of integration which allowed for instructing alexa

[wary trout]

Yes it's definitely possible. I did it. I don't remember though how - it was a year ago, since then i'm alexa-less for 7 months 😦

[opal zephyr]

I bet that feels great! 😄

[wary trout]

It wasn't that great at the start, since there was no good open source alternative. Now there is, and till the end of year there will be another two 🙂

[opal zephyr]

i'm getting actually quite close to bodging all of this if i can get either this or the spotcast integration to work with voice

which i've asked in (yet another) thread

[wary trout]

Yeah at some point i made choice between Google in Alexa to latter (it was better at voice), but then i realized that Alexa is pretty bad with HA, and very strictly locked...

[opal zephyr]

i think the problem is that i'm not sure if you can or if you should be able to change the pipeline at all so it should consider certain things when you ask ceertain things of it. ie does it know to use spotcast if i ask it to play music... does it know how to ask alexa something if i ask it to do something to alexa. etc etc

or "should it just work"

[wary trout]

It doesn't know anything, you have to teach it

[opal zephyr]

how though? 😦

through automations?

[wary trout]

Through custom intents

https://www.home-assistant.io/voice_control/custom_sentences/

https://www.home-assistant.io/integrations/intent_script/

[opal zephyr]

i keep hearing that term, but i've not found any reference to it

ah ok

so its all config.yaml stuff?

[wary trout]

Intent is

First link shows how to do some basic stuff with automation

[opal zephyr]

so intents are vague instructions still ? ie its not going to expect the exact statement of "can you do the washing up" and it will run whatever it is... it will kinda know if you say "i need my dishes washed" would mean the same thing and do it?

am i thinking along the right lines?

[shrewd veldt]

intents are rigid, meaning it has to exactly match the sentences as you define them. But you can define complex sentence structure to cover most utterances.

For something as flexible as you are asking (basically uttering something that is along the idea without using exactly defined words/sentences), you would probably need to go with LLM and expose tools to it, but that introduces other caveats and complexities 😄

[wary trout]

Well, for playing there could be several sentences added. But in the easiest case i'd just redirect full "play {play_request}" sentence to Alexa. Like, if sentence starts with "play" (and, say, finishes with "spotify") - just take full sentence and use it as command for Alexa (which is done with Alexa Media Player integration).

I use Music Assistant, but i don't use their own LLM integration, cuz it's bad. I use combination of my own intent and LLM to define exact track/album/artist and play it on MusicAssistant.

[opal zephyr]

could you possibly please give me an example of your intent config for doing that, maybe i can bastardize it for use with spotcast

[wary trout]

It's pretty extensive stuff, with several scripts and LLM connected. You sure you want it?

[opal zephyr]

hah well if it's ok with you , it will at least let me know what I'm getting myself in to ;-). but if it's a faff then don't worry

[wary trout]

Nah i'll pastebin some

1. So it all starts here, with custom intent script:
   https://pastebin.com/z9qEbscM - this is sentence part
   https://pastebin.com/Takr9Dfk - this is the intent script itself.
   All of this could be done in automation, i guess. It's just i'm using intent scripts for everything Assist-related.

2. Script to start work on request. It defines if play music or video. Latter isn't actually working well with Jellyfin (there's bug on Jellyfin), but i just don't use that part, didn't delete it so far. Don't mind it. 🙂
   https://pastebin.com/jRtibcfd

That script is launching "refine" work, and after that is launching playback on MA.

3. Refining user request. Basically i'm using GPT to pull data from user's request (this way i can play by request "play the track from Kill Bill")

https://pastebin.com/aT6gvhCV

4. And lastly, script to play it on MASS:
   https://pastebin.com/LvVRf0ZL
   That thing is searching media ID with mass.search, and plays it on MASS player in given area.

Uhh, just realized that with new Music assistant inbuilt integration "mass" services and integration are obsolete.

[wary trout]

It will still work with HACS MASS. For inbuilt it should be a bit adapted, but made it working too.