Can you give more details about the sliding window size?

[elephantpanda]

Hi I am trying to implement this in Unity.

I am trying to find out what is the sliding window size and how much it is slid each step?

Presumably each window embedding outputs a vector of size (1,1,1,1,96)

Then you take several of these vectors and use a to try to detect a word with input something like: (1,22,96)

Is that correct? I have found that an audio sample size of 12600 gives the correct size but not sure about how much the window increments each step.

[dscripka]

That is essentially correct, yes. The main audio embedding model (trained by Google) expects a time step of 80 ms (at 16 khz), so in general that is what I recommend. You can go lower or higher than this, but performance may decrease the farther away you get from 80 ms.