Excellent work.
Thank you, I was looking for something like this.
My daughter has a high pitch voice, and she can’t trigger the VA with any of the wake-words, I’m going to try to train a model using her voice to see if I can make her happy.
Hi , I created a custom wake word for myself, and it works fine. However, whenever I say the wake word, there’s always a delay of about 0.5 seconds before the assistant starts. If I use “okay_nabu,” the assistant starts almost instantly with no noticeable delay. I trained it using my own voice samples—how can I make my assistant launch as quickly as “okay_nabu”?
INFO ESPHome 2026.2.4
INFO Reading configuration /config/esphome/home-assistant-voice-094bce.yaml...
INFO Updating https://github.com/esphome/home-assistant-voice-pe@dev
INFO Updating https://github.com/esphome/esphome@b49b09b6ae56502aa3ce51be86f90d732d019b2c
WARNING GPIO0 is a strapping PIN and should only be used for I/O with care.
Attaching external pullup/down resistors to strapping pins can cause unexpected failures.
See https://esphome.io/guides/faq/#why-am-i-getting-a-warning-about-strapping-pins
WARNING GPIO3 is a strapping PIN and should only be used for I/O with care.
Attaching external pullup/down resistors to strapping pins can cause unexpected failures.
See https://esphome.io/guides/faq/#why-am-i-getting-a-warning-about-strapping-pins
WARNING GPIO45 is a strapping PIN and should only be used for I/O with care.
Attaching external pullup/down resistors to strapping pins can cause unexpected failures.
See https://esphome.io/guides/faq/#why-am-i-getting-a-warning-about-strapping-pins
Failed config
micro_wake_word: [source /data/packages/dc43710d/home-assistant-voice.yaml:1706]
id: mww
microphone:
microphone: i2s_mics
channels: 1
gain_factor: 4
stop_after_detection: False
models:
- model: |-
https://github.com/kahrendt/microWakeWord/releases/download/okay_nabu_20241226.3/okay_nabu.json
id: okay_nabu
- model: https://github.com/kahrendt/microWakeWord/releases/download/stop/stop.json
id: stop
internal: True
-
Not a valid model name, local path, http(s) url, or github shorthand.
model: |-
https://github.com/TaterTotterson/microWakeWords/blob/628bc036208883bdb725b2eda606a47f7f41bb85/microWakeWords/hey_sarah.json
id: hey_sarah
I mean, this was copy paste from your samples. Just replacing the URL to the JSON file and the “id” to hey_sarah
Okaaayyyyy… Thanks for that one. It did the trick; compiled and installed on PE device.
The device is back online and listens to “Okay NABU” So far so good. It’s not bricked at least
But my 4th MWW “Hey Sarah” does not appear in the dropdown of the ESPHhome device.
Only the standard ones. Nabu, Jarvis & Mycroft.
As far as I know, I have to insert the entire YAML code into esphome. However, this is problematic if something has changed. Then you have to search for your changes again. The above workaround leaves everything at default.
Hello everyone! I’d love to get some help setting up custom wake words, I’ve tried all the YAML configurations in this thread and even tried training my own model, however no mater what I do I really cannot get any of them to work.
I have cleared the cache and build files before every build and reset the VPE as described in the documentation; I’d love to at least get Alexa to work as I’m not English and the initial wake words are quite awkward in any other language.
I also tried referencing the Alexa model directly through microWakeWord, along with the internal stop model, as suggested earlier in the thread but alas to no avail.
Don’t use the “internal” key for the main models, as it will prevent you from seeing them in the interface and assigning them to a satellite.
“Stop” is a service ww, and its activation and deactivation occurs automatically, so we’re removing it from the visible list.
I’ve been using the stock “Hey Jarvis” but it struggles to pickup my wife and daughter. Before I realized I couldn’t perhaps easily replace the stock Hey Jarvis wake word I custom trained on our family’s voices for Hey Jarvis. Now when I adopt the Home Assistant Preview Edition into ESPHome it shows two Hey Jarvis wake words. In the ESPHome integration for the Voice Preview Edition it shows just one Hey Jarvis wake word. I’m not sure if that means both models will be used at the same time and what problems that might cause. Is there a way for me to remove the stock Hey Jarvis and just use my custom one?
