Hi Everyone ! I have been trying to setup the custom wake word for my Home assistant voice preview for a few days now. Is it possible? I have installed openwakeword and the supporting integrations successfully. I created a new pipeline. I clicked on the three dots in the upper right corner and clicked on streaming wakeword. I use HA Cloud for TTS STT and at the very bottom I clicked on the openwakeword engine and selected my custom wake word. For testing purposes I just selected Megan from a list of about 50 names. This tells me that the share that was created early on is working : ) I also confirmed the naming convention/formating of the file name is correct. I click update and all is good there. Then I go to the device page of the voice assistant and I select the pipeline with my custom wakeword and then click the drop down to assistant 1 and I only see the 3 default names, Jarvis , Mycroft, and Okay Nabu. No Megan??? Is there suppose to be another field or option on the device page to enable streaming wake word there too? In my trouble shooting of all this I’m hearing mention of this addition open rather it be a toggle switch or dropdown in the configuration section. I do not have that option, there is nothing like that on that page. I have the latest version of home assistant as of today and my firmware on the Voice assistant preview is Firmware: 26.4.0 (ESPHome 2026.3.2) I’ve also made sure everything is exposed. Is this not possible yet or am I missing some steps? I hear some people are doing it so I’m wondering if I just missed something. Thank you in advance if you can shed any light on this for me 
Voice PE uses on-device MicroWakeWord, not OpenWakeWord on your HA server. You need to train a MicroWakeWord model, and then flash that as part of a custom firmware build to the device.
Check out this project for MicroWakeWord training.
Hi,
Well I finally got Voice PE working with a custom wakeword.
Here are the steps I took…
Firstly a big shout out to Tater Totterson who will kindly create microwakeword models for you if you follow his instructions and create an issue in the required format on his github here…
TaterTotterson/Tater-Wake-Words
This created three files, two the yourwakewrord.tflite and yourwakeword.esphome.json are needed. I downloaded these to upload to HA, though you don’t actually need to do this as you can use the url to the files.
Next you need to add the ESP Home Builder app to HA. This runs as a separate app and in the dashboard it should discover Voice Preview. You need to take control of it.
I was a little reticent, however you can reflash if necessary using the Voice preview imager and plugging it into a laptop. I only had to do this once!
I’m running on HA Green and it wasn’t powerful enough to build an image, so I downloaded the ESP Home Builder windows version to my laptop and installed this.
There’s then a bit of faffing where you pair the laptop ESP Home Builder with ESP Home Builder App Dashboard running on HA. I eventually worked this out.
Using File Editor I uploaded in my case the hey_alan.tflite and hey_alan.esphome.json files to Home Assistant Green. I stuck them in the ESPHome directory. You don’t actually need to do this, you can put the url to the github in as it’s only used during the build.
Then in ESP Home Builder I clicked on core configuration and between api and wifi I added
micro_wake_word:
id: mww
models:
- model: /config/esphome/hey_alan.esphome.json
id: hey_alan
and save
The reason I needed to reflash the first time is there appears to be a bug in ESP Home Builder.
Clicking on wifi the SSID and password were for some reason set to my HA Green password not my wifi network name and password. Hence the first time I built an image and installed it voice preview stuck at the all white led boot stage.
So thinking this seemed to be the issue I changed the SSID and password to my wifi network. I got a warning about secrets.yaml and updating everytwhere.
I then clicked on the three … and validated the image
Then I click on the install arrow.
The build kicks off on the laptop and initially took quarter of an hour. If you’re on a more powerful host than HA green then building locally might be possible.
This time the image installed over the network and voice PE came to life.
It reappeared in the integrations discovered window and I clicked on reconfigure.
It wanted me to enter a key and this was found on the ESP Home Builder dashboard three dots - show api key.
I then went into integrations ESP Home and re set up the voice assistant - this time my custom wakeword ’ Hey Alan’ is in the list and it works pretty well. Dance around the room.
I hope this helps. Everything that should seem obvious, isn’t the first time you do it.
If you do trash the image and get stuck, go to nabu casa and follow the reflash image instructions here Reinstalling the firmware on Home Assistant Voice Preview Edition – Nabu Casa Support
It’s quite straightfoward with a laptop and usb cable
I added the Voice Preview device to ESP Home Builder and it seems that it freaks out because the firmware on github is no longer avaiable. Let me know what you did to get around that issue. I see on google AI results that they do not advise you “take control” of the device anymore because of this issue. Here is an image of what I am seeing after I take control.
Github is sick today, don’t read anything about the software into those errors.