DTW를 이용한 선박 음성조타명령 인식시스템의 구현

Abstract

In this dissertation, we present an implementation of automatic speech recognition system designed to recognize vessel's steering command. The system first detects the point where the speech starts and ends, and then processes the acoustic signal to produce 13th order MFCC. The recognition of steering command is conducted in 2 steps: the calculation of similarity in word level, and then the selection of the final steering command from the set of feasible commands. The word level similarity between test and reference MFCC vectors is calculated by DTW (Dynamic Time Warping) method. Then a word lattice is made to aid the selection of the steering command. Steering commands considered in this study consist of one to four words chosen from 23 words including numbers. Three kinds of recognition experiments are conducted to test the performance of the presented system. In the first two cases which are speaker-independent command-level and word-level experiments, the recognition rate was around 70%. In the last experiment, which is speaker-dependent command-level recognition, the recognition rate is more than 97%, which encourages the application of the system in real environment.