Assist no longer requires wakeword. If you want that

Configuration Voice Assistant
1

Unfortunately after the voice contest.

I can now activate and deactivate the wakeword and activate the listening manually after the sentence from jarvis is finished.
So you can now make a normal conversation with youre assist. Its the first version and needs a little updates. But i am very happy that this is working.

I have an ESP32 as microphone, HA, chatgpt and an echo.dot as loudspeaker.

The ESP recognizes the wakeword the first time, but after that the wakeword is no longer needed.

The ESP records what is spoken, then it is converted to text and sent to gpt which sends back its response and HA makes a media file mp3 which is sent to Alexa for playback. So far the normal state.

How does it work now?
The mp3 is stored locally on the system, this URL is tapped and when the URL is created it is sent to a pyscript. This opens the file, reads the mp3 duration, adds more seconds depending on the media length and writes them to a timer. After the timer has expired, listening is reactivated.

sorry the first video was a earlier not finished version. now its the correct version.

2
3

Are you using conversation.process to interact with gpt?

4

can you explain what you mean? i am using the normal way from esp to HA to the gpt integration.

5

I am using a modified wyoming satellite branch to activate my raspberry satellite without trigger word but I am working in some way to have continuous conversation in a correct way (with node red) but I have not found a reliable flow…Currently I am trying using che conversation.process action of home assistant to interact with chatgpt…

7

here you can find the manual i made for this. GitHub - Coketrd/NoWakeWord_DoAConversation: Do a Conversation with youre HomeAssistant-Assist without needing a WakeWord the hole time.

8

I am using little bit different approach for continuous conversation

  1. first wakeword runs continuous conversation (if selected on ESP32 switch)
  2. automation disables listening on ESP32 mic when media_player is playing
  3. after media_player ends the play (TTS MP3) mic is enabled for 30s
  4. when there is STT detected during those 30s, conversation continues
  5. when tehre is 30s silence, listening ends and waits for wakeword.
9

The echo.dot didnt send the status mp3 playing and ending. It only does if you play musik. But the playing mp3 for answering didnt change the status in HA. Because of this i check the duration from the mp3 files. Maybe the google does this different. This was my first plan too to activate after stop playing.

10

how you let the mic 30s open?

12