Add tts-finished state to simple_state sensor, continue_conversation switch and custom_responses_blueprint

[relust]

Hello. Congratulations on what you've been able to do with this integration. In order to be able to properly implement Hassmic integration in the ViewAssist project, I made some changes to the code.

1. Tts-listening state implementation
   One of the things needed to use this integration is to be as precise as possible when the tts message actually finishes playing because TTS_END marks when the message is sent to the player and not when it finishes playing. That's why I took this implementation from the StreamAssist integration and looked for the best way to implement it in Hassmic. Practically, when the TTS_END event is executed, the URL of the tts is taken and using mutagen.mp3 to calculate the effective duration of the message. Then in the SimpeState sensor I put a new state tts-finished.
2. Correcting the publication order of stt-speaking, stt-listening, wake_word-listening/stt-listening statuses
   There is a problem though with the fact that the wake_word-listening state fires before the message finishes playing and at the end it remains as the last state tts-finished and not wake_word-listening. That's why I modified the code to respect the order of publishing the statuses: stt-speaking, stt-listeningand at the endwake_word-listening or stt-listening, depending on how the continue conversation switch is set. Practically, I added a delay in the publication of wake_word-listening or stt-listening statuses until the tts message is finished playing. This does not affect the functionality of the satellite in any way because the Simple state sensor is just a sensor made especially for the View Assist project. Listening to the wake word starts immediately after TTS-END, but in order for the avatar to remain talking until the end of the message playback, this is necessary.
3. Continue conversation switch
   I also added a switch to enable continue conversation. Changes must also be made in the _init_.py file from the switch directory, in the pipeline_manager.py file and add continue_conversation.py . To be able to use continuous conversation in Hassmic, a switch can be used to choose where to start the pipeline, from WAKE-WORD or from STT-START. In the continue_conversatin switch code, you can set how many seconds to wait for the next command until it switches back to wake word detection. From what I noticed, after 9 seconds it is no longer in the listening state and the pipeline resumes from the start.

            done, pending = await asyncio.wait(
                [vad_task, error_task],
                timeout=9,
                return_when=asyncio.FIRST_COMPLETED,
            )

Because stt-listening starts before the tts message is finished playing, I made an automation in the continue_conversation switch code that mutes the microphone between the tts-speaking and tts-listening statuses of the simple_state sensor. Because I failed to import the simple_state sensor internally in the continue_conversation switch, I chose to import it from the home assistant, but I think it would work better if you managed to import it directly from the hassmic integration.

4. Conversation_id
   In pipeline_manager I put conversation_id=”123456789” and it works fine with OpenAi to retain conversation history for last 20 messages. Maybe we will think of a logic to generate random ids for conversation_id but keep a specific Id in certain situations if we want to retain the history of the conversation. I believe that the satellite, as in our Hassmic case, is the one that generates the conversation_id and the Home Assistant pipeline takes the conversation_id from the satellite.

                    ),
                   start_stage=start_stage,  # It uses dynamic logic for the starting stage
                   device_id=self._device.id,
                   conversation_id="123456789",  # Or get this value from a dynamic source
               )

5. Custom responses blueprint
   Optionally you can use uith that the custom_responses blueprint which gives an answer after detecting the wake word (Ex: yes i'm listening) or if you say ok nabu twice or when it is in continue_conversation mode and you don't realize that the satellite is actually waiting for the command and not the wake word.

blueprint:
  name: Viewassist Custom Responses
  description: Automation for custom responses in Viewassist with selectable media content ID, media player, microphone, and custom wake words.
  domain: automation
  input:
    trigger_entity:
      name: Trigger Entity
      description: Microphone entity sensor of satellite ex. sensor.viewassist_office_simple_state
      selector:
        entity:
          domain: sensor
    media_player:
      name: Media Player
      description: Select the media player to use
      selector:
        entity:
          domain: media_player
    microphone_switch:
      name: Microphone Switch
      description: Select the microphone switch to turn off and on
      selector:
        entity:
          domain: switch
    media_content_id:
      name: Media Content ID
      description: The media content ID to play (optional, with a default message)
      default: "media-source://tts/edge_tts?message={{ ['how can I help you', 'yes, i`m listening', 'how can assist you'] | random }}&language=en-US-ChristopherNeural"
      selector:
        text: {}
    custom_wake_words:
      name: Custom Wake Words
      description: Add custom wake words (e.g., 'ok nabu'). Add multiple values if needed.
      default: ["ok nabu", "ok naboo", "okay naboo"]
      selector:
        object: {}
    conversation_response:
      name: Set Conversation Response
      description: The response text for custom wake words
      default: "say please"
      selector:
        text: {}

trigger:
  - platform: state
    entity_id: !input "trigger_entity"
    to: wake_word-detected
    id: wake_word_detected
  - platform: conversation
    command: !input "custom_wake_words"
    id: custom_wake_word_detected

condition: []

action:
  - choose:
      - conditions:
          - condition: trigger
            id: wake_word_detected
        sequence:
          - service: switch.turn_off
            target:
              entity_id: !input "microphone_switch"
          - service: media_player.play_media
            target:
              entity_id: !input "media_player"
            data:
              media_content_id: !input "media_content_id"
              media_content_type: provider
              announce: true
          - wait_for_trigger:
              - platform: state
                entity_id: !input "media_player"
                to: idle
            timeout:
              hours: 0
              minutes: 0
              seconds: 1
              milliseconds: 500
          - service: switch.turn_on
            target:
              entity_id: !input "microphone_switch"
      - conditions:
          - condition: trigger
            id: custom_wake_word_detected
        sequence:
          - service: media_player.play_media
            target:
              entity_id: !input "media_player"
            data:
              media_content_id: !input "media_content_id"
              media_content_type: provider
              announce: true
          - set_conversation_response: !input "conversation_response"
            enabled: true

mode: single