Can speech recognition be encoded directly

We add the sound data directly every m data, and use s to represent the value of m data, the sound becomes a square wave

Next, we divide the y value into two regions
that are greater than the median value m and less than the median value
. Value range is
$(max-m)/N,(m-min)/N$
a value v within a certain range of values ​​is taken as the minimum value in the range,
after comparison, the size of the value is not looked at, as shown in
thisfigure, speech recognition can be performed