The DTWA-Based Real-Time Digital Speech Recognition System

• Main Authors: Klim Pan, 潘克寧
• Other Authors: Chen-Mie Wu
• Format: Others
• Language: zh-TW
• Published: 1995
• Online Access: http://ndltd.ncl.edu.tw/handle/44784312330282137032

碩士 === 國立臺灣科技大學 === 工程技術研究所 === 83 === In this thesis we present a real-time digital speech recognition system which consisters of a PC-486, a Dynamic Time-Warping Algorithm(DTWA) processor, a Sound Blaster card, and a microphone. Also, the system presented can be separated into the following four parts: speech sampling, preprocessing, template matching, and postprocssing. First of all, the analog speech signal is sampled by using microphone and Sound Blaster card, and, stored into the memory of PC-486 in digized form. Secondly, preprocessing program running on PC-486 extracts feature vectors from the sampled speech signal to form an input template. Then, under the control of PC-486, the DTWA processor executes template matching between input and reference templates. Finall, the postprocessing program running on PC-486 obtains the recognition results through analyzing the matching results computed by the DTWA processor. Currently, the DTWA processor we design can execute more than 10,000 template matching per second. With such a performance, the digital speech recognition system presented in this thesis has satistfied the real-time requirement. Additionally, multiple-reference-template and unconstrained-endoint template matching methods can be applied easily to enhance the recognition rate.