Experiments on the Use of LVQ in Phoneme-Level Segmentation of Speech