Speaker-dependent Isolated-Word Speech Recognition System Based on Vector Quantization

Speaker-dependent speech recognition system requires the system should not only recognize speech, but also recognize the speaker of the segment. In this paper, two indicators are selected—short-time average zero-crossing rate and dual-threshold endpoint to test the signal endpoint through the study of speaker-dependent isolated-word speech characteristics, and MFCC parameters are taken as the characteristic parameters; based on vector quantization, template matching algorithms are designed, and one is adopted to improve LBG algorithm to increase the computing speed; speaker-dependent isolated-word speech recognition system is designed based on vector quantization technique and simulation experiments are conducted in the MATLAB platform under various backgrounds, which proves the system has better recognition effect.