Larger window to summarize features
To model the time it takes a human to "understand" and extract texture information from a signal
Treat points within the "texture window" as if points in a Gaussian distribution
Then get mean and variance
"Another way to think of the texture window is as a memory of the past"
"In our system, an analysis window of 23 ms (512 samples at 22 050 Hz sampling rate) and a texture window of 1 s (43 analysis windows) is used."