A New Spectral Normalization Technique for Speaker Independent Speech Recognition