How to use Vosk to realize voice wake-up function？

[WENxxxXXX]

I'd like to use Vosk for voice wakeup in my browser, but I'm running into issues with false wakeups and difficulty getting the system to wake up.
I've read the documentation on fine-tuning Vosk: https://alphacephei.com/vosk/lm, but I'm still not sure how to improve the model's wakeup word recognition accuracy.Could you please provide more detailed information about the modification of information sources in Vosk model? Thank you very much!

[WENxxxXXX]

I used this model: vosk-model-small-cn-0.22.

[nshmyrev]

You'd better use some specialized model. For Chinese you can try

https://k2-fsa.github.io/sherpa/onnx/kws/index.html

[WENxxxXXX]

Thank you very much. I'll look into it

[bhu1-103]

hey @WENxxxXXX , you can try this out

1. Transcribe everything continuously.
2. Pass the transcribed text over a fuzzy matcher (e.g rapidfuzz).
3. If it matches your wake word, by let's say 80% or so, then trigger your wake up action.

This way, you dont need to modify the code of the Vosk model itself.
You can also tweak the threshold to reduce false positives.

check out how i used it in my own code here