Behaviour after running the model

Help
#1

Hi,
I’m using the audio classification on Raspberry Pi 4 and I have two questions:
1- Why does the model detects the last class (with .99) when I first start running the model and there is silence and still I haven’t given any sound?

2- I edited the code so the model print the name of the label predicted. However, when I run the sound once, it prints multiple times the label detected. why is it not printing it only once?

Thanks

#2

Hi @Aisha,

  1. What does your last class correspond to? A Neural Network always predicts an output, you could create a silence class and filter it out if you don’t want it to be displayed.
  2. If your sound is longer than the window size, it will be displayed multiple times as we perform continuous inferencing, see more details in the code here.

Aurelien

#3

Thanks @aurel . So does it mean the second or last result is the most accurate? how can I just get one result from one voice input? My window size is 2000, the model was trained on audio with 2500 ms.