Constructing Invariant Representation Of Sound Using Optimal Features And Sound Statistics Adaptation