한국해양대학교

Detailed Information

Metadata Downloads

ART2 적용 임베디드 음성인식 시스템의 설계 및 구현에 관한 연구

Title
ART2 적용 임베디드 음성인식 시스템의 설계 및 구현에 관한 연구
Alternative Title
A study on the Design and Implementation of ART2 Application Embdded Speech Recognition System
Author(s)
류홍석
Publication Year
2004
Publisher
한국해양대학교 대학원
URI
http://kmou.dcollection.net/jsp/common/DcLoOrgPer.jsp?sItemId=000002174179
http://repository.kmou.ac.kr/handle/2014.oak/8133
Abstract
Speech recognition called machine receives human's language and achieves suitable action according to this language. This technology is used on industry whole, and is specially applied to information industry field, digital communication, electronic, multi media etc. Apply to mobile robot that is electric motion wheel chair system at LAB through this technology. So developed to give more convenience for hand and feet uncomfortable disabled person.

In this study, consider that the plant is electric motion wheel chair. So during speech recognition, is used DTW(Dynamic Time Warping) that relative correct recognition rate is fine being speaker dependent type. But consider that real-time have to get small memory and fast processing speed. So introduced VQ(Vector Quantization) used in data compression algorithm of speaker independent. In accordance with, secure fast recognition and small memory. But discovered that recognition rate is fallen by using VQ. So, in after treatment algorithm for correct recognition rate enhancement ART2(Adaptive Reason Theory 2) algorithm application on about 5% correct recognition rate enhancement bring. To use ART2, must be applied error range. Er ror range is applied result that extract 1 order distance in 2 order distance is more than 20 by each distance to apply DTW. Like this, bring fast processing and high correct recognition rate apply ART2.

Because it is moved object, must implement by embedded system. So It choose chip of TMS320C32 that processing a lot of computation complexity relatively fast and implement embedded system. Memory can store a lot of data in speech considering, therefore possessed 128kbyte's RAM memory and 64kbyte's ROM memory. Input of speech use 16bits stereo audio codec, secure relative correct data through high resolution. The mobile robot uses 80C196KC. and output PWM generating power through HSO that chip had and designed to run motor.
Appears in Collections:
전자통신공학과 > Thesis
Files in This Item:
000002174179.pdf Download

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.

Browse