음성인식 기술을 활용한 영어 발음 자동 평가 시스템 및 방법에 관한 연구

국문 초록 (Abstract)

본 연구는 영어 발음 자동 평가 시스템 및 방법에 관한 것으로, 피평가자
에게 문제를 제공하는 문제 제공부, 상기 문제에 따른 피평가자의 발화 음
성을 입력받는 발화 음성 입력부, 음성인식 기술기반으로 상기 발화 음성을
음성열로 변환한 발화 정보를 생성하는 발화 정보 생성부, 상기 발화 정보
로부터 상기 발화 음성의 높낮이를 표시한 억양 정보를 추출하고, 상기 억
양 정보를 기반으로 상기 발화 정보가 평가된 제1 평가 정보를 생성하는 제
1 평가부, 상기 제1 평가 정보로부터 상기 발화 음성의 길이를 추출하고, 상
기 발화 음성의 길이를 기반으로 상기 제1 평가 정보가 평가된 제2 평가 정
보를 생성하는 제2 평가부, 상기 제2 평가 정보로부터 의미상 독립적 경계
유무를 판별하고 상기 제2 평가 정보에 상기 의미상 독립적 경계 당 하나의
휴지구간(pause)이 추가된 제3 평가 정보를 생성하는 제3 평가부, 상기 제3
평가 정보를 점수화하고, 피평가자에게 평가점수를 제공하는 평가점수 제공
부 및 상기 문제 또는 평가점수 중 적어도 하나가 표시되는 인터페이스부를
포함하는 영어 발음 자동 평가 시스템 및 방법에 관한 연구이다.

다국어 초록 (Multilingual Abstract)

This paper aims to describe an automatic English pronunciation
evaluation system and method using deep learning. The system consists
of a problem provider that provides problems to an evaluated person, a
speech input unit that receives speech from the evaluated person, a
speech information unit that converts speech into mechanical sound
energy and generates speech information using the voice recognition
technology. This study looks into a procedure used for an automatic
English pronunciation evaluation method, which consists of the 1st
evaluation unit that generates an intonation information indicating the
height of speech extracted from the above speech information, the 2nd
evaluation unit that determines where there is an independent semantic
boundary using the above intonation information the 3rd evaluation unit
that adds numbers of pause information, based on the above semantic
boundary information. This research identifies the optimal algorithm for
an automatic English pronunciation evaluation showing that the 3rd
evaluation unit scores the 3rd evaluation information and the system
displays the 3rd evaluation information at a problem provider that
provides evaluation scores to the evaluated person and at an interface
unit that displays either problems or evaluation scores.

목차 (Table of Contents)

• 제 1 장 서론 ···························································································1
• 제 1 절 연구 배경 ·····················································································1
• 1. 기술 분야 ·····························································································1
• 2. 연구의 배경이 되는 기술 ········································································1
• 제 2절 연구방법 및 논문 구성 ····································································2
• 제 2장 이론적 배경 ····················································································4
• 제 1 절 딥 러닝 기반의 음성인식 ································································4
• 1. 딥 러닝의 정의 ······················································································4
• 2. 딥 러닝 기반의 음성인식 ········································································4
• 제 3 장 본론 ·····························································································5
• 제 1 절 영어 발음 자동 평가 시스템 ···························································5
• 1. 전체 아키텍처 ·······················································································5
• 2. 원어민 발음 유사 정도 판단 모델 ···························································7
• 제 2 절 음성인식 기술 기반 영어 발음 평가 방법 ········································16
• 1. 데이터 소스 ·························································································16
• 2. 영어 발음 평가 지표 ·············································································17
• 제 4 장 실험 및 분석 ················································································18
• 제 1 절 실험 환경 및 방법 ········································································18
• 1. 영어 발음 평가 실험 환경 및 방법 ··························································18
• 2. 인간 교습자에 의한 영어 발음 평가 ·························································19
• 3. 음성인식을 활용한 영어 발음 자동 평가 ···················································20
• 4. 평가점수 제공 모델 ················································································20
• 제 2 절 결과 및 분석 ·················································································23
• 1. 결과 ······································································································23
• 2. 분석 ······································································································23
• 제 5 장 논의 및 결론 ·················································································25
• 제 1 절 연구 요약 ······················································································25
• 제 2 절 연구의 시사점 및 향후 연구 방향 ···················································25
• 1. 연구의 시사점 ························································································25
• 2. 연구의 한계점 및 향후 연구 방향 ····························································25
• 참고문헌 ·····································································································27
• Abstract ····································································································29