Parallel notifications

[mkrusemark]

Is your feature request related to a problem? Please describe.
Notifications sent to multiple devices at home are sent in sequence. In consequence one speaker after the other uses tts to say the message.

Describe the solution you'd like
Parallel execution of notifications of all targets.

Describe alternatives you've considered
Single notify.supernotify actions. But this would make this great integration almost useless.

Additional context
Think about a delivery like this:

speakers:
      transport: tts
      selection: scenario
      options:
        unique_targets: true        
        target_usage: fixed
      target:
          - media_player.echo_g
          - media_player.echo_f
          - media_player.echo_bu
          - media_player.michael_s_sonos_beam
          - media_player.home_assistant_voice_09854a_media_player

[jeyrb]

Have noticed that myself with the Alexa Devices integration, when I give it a list of entities and the bedroom is noticeably later than the kitchen. Supernotify itself isn't blocking parallelisation, since it

I think reason for this is within the media_player integration, which in turn is called by the tts integration. Since its an abstract integration its the entity_platform which distributes out the calls (https://github.com/home-assistant/core/blob/f06f25b99ab182145815f6fd96d260cc502184c8/homeassistant/helpers/service.py#L725) - it does say its simultaneous but at the level of entity, not at the level of platform. So my reading is your targets would turn into 5 entity calls rather than 1 call for Amazon Echo, 1 for Sonos and 1 for HA Voice

Only common solutions for simultaneous announcing I've seen are Sonos and Alexa Devices, where for the latter you can create a Group within Alexa app, which appears to home assistant as a single device.

It might be tricky to bypass the entity platform handling, since media_player concrete devices get mostly hidden completely behind the media_player abstract platform. I'll have a look, since if can get platform level calls with multiple entities per call, then could take advantage of platforms that can send out simultaneously .

One thing that might improve things slightly is to change the order of entities, since the Entity Platform code respects order, and so if the first entity is much slower it will slow down the rest (the AI work for actual TTS should already have happened just once for all the entities prior to that )