#Wake word on device vs custom wake words
1 messages · Page 1 of 1 (latest)
by default, currently yes.
however you can add some extra to the firmware if you want to customise it.
in the near future supporting externally loaded MWW models from the home assistant server will be possible. the feature backlog issue is HERE support is being added with some prs in both HA and ESPhome. and its coming along.
it should also be noted that MWW models are not the same as OWW models and they are not compatible. making a MWW model is very much not a simple task. there are a few custom ones around which have varying degrees of results.
making a wake word model is very challenging and requires a bunch of real world sample data.
So you are saying the whole huge collection of OWW models amassed by people is now useless?
you can in theory still customise voice satellites firmware to use OWW as a "streaming wake word" instead if you wanted to but... yes MWW is a different and more efficient system and not compatible with OWW models.
OWW is on-server wake word. It's much more efficient to keep wake word on the satellite though.
There are ports of OWW to Android, for instance, and you can use that huge list there. But for esp32 OWW is the heavy.
I have custom microwakeword running on two ESP32 S3 Box3 satellites. Answers to “Rosita” works pretty well. Did it on my MacBook Air, she needs some fine tuning and I want to make better models but so far she’s pretty solid.
Voice Assistant named Rosita.
• ESP32-S3-BOX-3 (from Espressif)
• Home Assistant Green
#HomeAssistant #ESP32 #ESPHome #MicroWakeWord #DIYAssistant #VoiceAssistant
So far all training software I saw works on Nvidia. What did you use for Apple chip?
P.S. is it Rosita from "Sing"? 🙂