Wake word detection with ESP32 satellite not working

Help
,
1

Hi
I’m trying to set up ESP32 as a satellite for rhasspy.

My setup:

  • ESP32 with inmp441 and button for hardware wake up. (SiteId: Test)
  • Raspberry Pi 3B hosting mqtt broker and rhasspy 2.5.11(in docker container). (SiteId: Master)

What works:

  • Audio from - hermes/audioServer/Test/audioFrame - loud and clear
  • triggering hardware wake up works perfectly - sesion starrted and intent detected

What does not work:

  • Detecting wake up word - I tried 2 different models with porcupine (jarvis and bumblebee) with every sensitivity level ; pocketsphinx with different key phrases and snowman. No session is started no intent detected.

What I already tried:

  • variety of models, sensitivities and methods of detecing wake word,
  • hosted rhasspy on different machine (x86)
  • tried rhasspy 2.5.10

rhasspy config :

{
    "dialogue": {
        "satellite_site_ids": "Test",
        "system": "rhasspy"
    },
    "handle": {
        "satellite_site_ids": "Test"
    },
    "intent": {
        "satellite_site_ids": "Test",
        "system": "fsticuffs"
    },
    "microphone": {
        "system": "hermes"
    },
    "mqtt": {
        "enabled": "true",
        "host": "192.168.40.2",
        "site_id": "Master"
    },
    "sounds": {
        "system": "hermes"
    },
    "speech_to_text": {
        "satellite_site_ids": "Test",
        "system": "kaldi"
    },
    "text_to_speech": {
        "satellite_site_ids": "Test",
        "system": "nanotts"
    },
    "wake": {
        "porcupine": {
            "keyword_path": "bumblebee_raspberry-pi.ppn"
        },
        "satellite_site_ids": "Test",
        "system": "porcupine"
    }
}

What am i doing wrong, what else can I try to make it work?

2

Maybe a dumb question but do you have a microphone attached to the ESP32?

3

I have inmp441 connected. And I know it works as I can clearly hear the audio which is published to hermes/audioServer/Test/audioFrame and also when I use button to trigger the intents are correctly recognized.

After some more fiddling around I found out that if I choose porcupine and americano model it triggers rhasspy every 5 or 6 times.