presented at event 2004 IEEE International Conference on Acoustics, Speech, and Signal Processing Conference