is there a way I can make Wake Word sensitivity on Voice PE even less sensitive

[velvet ridge]

I already have the voice PE set to slightly sensitive. Is there something i can add to the yaml code to add more, even less sensitive options? my custom wake word keeps triggering from the TV constantly

[desert rivet]

What is the wake word? How many syllables is it?

[velvet ridge]

its a custom wake word, "Echo". but the things that trigger it arent even close to echo

half the time "ok google" triggers it

[desert rivet]

So yeah it is pretty short, A two syllable wake word is probably going to get quite a lot of false positives.

[velvet ridge]

sure, but is there not a way to make the wake word even less sensitive than the most sensitive option?

[desert rivet]

you can try to manually set the probability cutoff higher and see if it helps.

[velvet ridge]

ok. how do i accomplish that?

[desert rivet]

can you paste your esphome code that adds the model?

[velvet ridge]

heres the full code:

  name: home-assistant-voice-0a020e
  friendly_name: Downstairs Basement Livingroom Voice PE
packages:
  Nabu Casa.Home Assistant Voice PE: github://esphome/home-assistant-voice-pe/home-assistant-voice.yaml
esphome:
  name: ${name}
  name_add_mac_suffix: false
  friendly_name: ${friendly_name}
api:
  encryption:
    key: Tnzed05IlUKxZiJRG/HbKfcdSdHMDp+ARr4wtBUIon0=


wifi:
  ssid: !secret wifi_ssid
  password: !secret wifi_password

micro_wake_word:
    models:
    - model: https://raw.githubusercontent.com/JohnnyPrimus/Custom_V2_MicroWakeWords/refs/heads/main/models/computer/computer.json
      id: computer
    - model: https://raw.githubusercontent.com/TaterTotterson/microWakeWords/refs/heads/main/microWakeWords/echo.json
      id: echo ```

[desert rivet]

I think you do it like this:

micro_wake_word:
    models:
    - model: https://raw.githubusercontent.com/JohnnyPrimus/Custom_V2_MicroWakeWords/refs/heads/main/models/computer/computer.json
      id: computer
    - model: https://raw.githubusercontent.com/TaterTotterson/microWakeWords/refs/heads/main/microWakeWords/echo.json
      id: echo
      probability_cutoff: 80

I have no idea what the default is tbh.

[velvet ridge]

ok, ill play with it, thanks

[desert rivet]

oh it might be 0.8 for 80% I am not sure.
looking at your model file it looks like the echo model default is 0.97.

{
  "type": "micro",
  "wake_word": "echo",
  "author": "Tater Totterson",
  "website": "https://github.com/TaterTotterson/microWakeWord-Trainer-AppleSilicon",
  "model": "echo.tflite",
  "trained_languages": [
    "en"
  ],
  "version": 2,
  "micro": {
    "probability_cutoff": 0.97,
    "sliding_window_size": 5,
    "feature_step_size": 10,
    "tensor_arena_size": 30000,
    "minimum_esphome_version": "2024.7.0"
  }
}

It is a float
probability_cutoff (float): The probability cutoff for the wake word detection. If the probability of the wake word is below this value, the wake word is not detected.
so probably is 0.97 for 97% within ESPHome config too
Documentation Link - https://esphome.io/components/micro_wake_word/

You might want to go with "hey echo" or "okay echo" or something though to add more syllables.

[velvet ridge]

ive been saying "echo" to trigger my voice assistants for 11 years now, cant go back

so add probability_cutoff: 0.80 to the bottom of the code and adjust the 0.80 as needed?

[desert rivet]

Yeah, Looks like it is probably something like:

    - model: https://raw.githubusercontent.com/TaterTotterson/microWakeWords/refs/heads/main/microWakeWords/echo.json
      id: echo
      probability_cutoff: 0.98

That you want.

Looks like the default for that model is 0.97 so so you want to go up from there.

[velvet ridge]

so far so good. everything updated fine and its responding to echo fine. no false triggers so far. i set the probability to 0.70

[velvet ridge]

welp, no joy. ive set it as low as 0.01 and its still false triggering. guess ill be stuck saying "computer" even tho i normally have to say it up to 5 times for it to register

[desert rivet]

You want to turn it up, Not down.

[velvet ridge]

oh, so if i want it less semsitive, i go up? like 2.50 or something?

[desert rivet]

It is percentage based so you can go over 1.

The model default was set to 0.97 so try 0.98 or 0.99.