Spontaneous unwanted reaction

I used a Raspberry Pi 3 Model B+ and a MakerHawk ReSpeaker 2-Mics Pi HAT to set up a snips system. The recognition rate is very good, and most of the time the system works as expected.
The Hot Word “Hey Snips” has not been changed.
But sometimes Snips thinks that Hot Word has been spoken. Afterwards an intent is apparently recognized and an action is triggered. The words spoken in the room are not similar to the hot word or the words of the intents.
Does anyone know the phenomenon?

Many greetings,

Yes happens a lot… I don’t use hey snips… try setting up your own custom wake word

Hello, ozie,
Is that really the case? Do you know more about the backgrounds?
If it were only the Hot Word, none of the programmed intents should be recognized and executed afterwards. That is very strange…
Thank you for your answer, I will try to set up my own Hot Word.

Many greetings,

Hi Erwin,

Have you tried to change the wake word sensitivity ? By default it is set to 0.5 but you can change that in the toml configuration file. You can decrease a lisstle the sensitivity and (say to 0.4) if needed.


Hi @Erwin,

Also, the current version of the hey snips hotword shipped in an assistant downloaded from the console is greatly improved and reach false positive immunity level you will clearly not reach with a custom hotword.

The more spoken utterances you will have next to your setup the more likely you have a risk of false triggering (let’s say the TV is running in the room, or many people are speaking in the room).

So yes, in this case you can try to lower the sensitivity level of the wake work detector. Check out the platform configuration documentation to do so.

Regarding the intent detection,

Once a wake word is detected, the ASR turn on and the NLU can possibly infers some false positive intents.

Many thanks for the advice!
I have adjusted the file “/etc/snips.toml” and the value “sensitivity” is set to 0.4.
Let’s see how the behavior will be in the next days…

Many greetings,