A speaker-independent isolated word recognition system is described which is based on the use of intra-word local and global structural features. The local features are incorporated into the system with using matrix quantization of segment patterns that are obtained by dividing a speech word pattern into partial patterns of equal frame length of about 4. Then global features are incorporated by making use of the correlations between segments. On a data base with a vocabulary of 212 words spoken by 20 speakers (10 male and 10 female), the system shows higher performance in recognition accuracy and processing speed than the system using whole-word template based dynamic time warping (DTW) algorithm.