fine-tuning of speaking-detection

We are getting info that the automatic attenuation of the floor audio is not working sometimes. We suspect that maybe the speakDetection is not triggering in some situations.

How should we change the setting so make it more sensitive -- make it attenuate more often?

I am referring to this setting threshold

    translation:
      origineVolume: 0.2
      translator:
        speakDetection:
          enabled: true
          timeout: 60000
          delay: 5000
          threshold: -70

or do you have other suggestions?