actually the idea, is to take speech
the first thing to do is post processing such as High pass filtering to remove noise and resampling to 8 kHz. Then each 4 samples will be compared to 4 parameters and calculate euclidien distance then element to element, the result ei=e1,...e4 is interpolated using Lagrange method. inputs ei and I=1,...4 and y=p(ei). the result will be auto-encoded to compress p(ei) and produce a compact representation.
I need math prove of it.