Why never "no intent recognized"?

1

Hi,

I am using Rhasspy 2.5 and have configured seven different sentences, which lead to a well defined action.

The problem is: I happens from time to time, that I have a false positive wake word recognition - I think this can not be avoided.

BUT: No matter, what bullshit I talk after the wakeword, Rhasspy ALWAYS uses randomly one of my preconfigured sentences. I tested with just nonsense “blablabla”… and still is matching to one of my sentences.

Question: is there any way to change the behaviour, that a sentence is only matching, when the correlation is really high? Otherwise it should lead to a “no intent recognized” and shall be ignored.

Thanky you very much!

2

Ah yes, false positives :frowning: Especially from background TV and conversation.

From memory, a previous suggestion was to add a sentence “nevermind” or similar, which does no action. Of course that will only work if you realise rhasspy was accidentally triggered and remember to tell it to “nevermind”.

I personally use Node-RED to process my automations for Home Assistant, and found there is a msg.payload.asrConfidence field which can be used to ignore commands with low confidence.

if (msg.payload.asrConfidence < 0.4) {
    msg.payload = '{ "sessionId": "' + msg.payload.sessionId + '", "siteId": "' + msg.payload.siteId + '", "text": "I didnt hear that clearly"}';
    msg.inSession = true;
    return msg;
}

The long-term answer is to improve the voice and wakeword recognition.

3

Any ideas about improving wakeword detection? I manually updated to latest Porcupine Version and generated some custom wake works, but this did not improve the situation

I dont see any asrConfidence in my Node-RED:

nodered
nodered752×1672 152 KB