using distance or distortion measures between unknown speech and reference templates