線上美式英文口音訓練系統

隨著網路英語教學的發展，國人越來越重視全面性的學習與能力訓練。從早期的托福測驗，到現今的全民英檢以及多益測驗都加入了口說能力的鑑定，以符合英語在生活中的實用性與溝通性，因此培養口說能力，已成為目前英語教學上的一個重點。當我們與外國母語人士做溝通時，由於對「發聲方式」不同認知，以致經常發出不正確的發音，造成溝通上的困難與誤解。我們的系統將致力於訓練學生的口說能力，透過我們所提供的錄音功能作反覆練習，從老師的批改以及系統的音高分析中糾正學生的發音。因此本篇論文運用「語音辨識」和「字串比對」等相關語音合成技術，藉由聲音辨識模組的結果，分析出各個單字的時間點，並透過語音對文字的交互比對(Speech-Text Alignment)和動態演算法的調整，完整地找出文字與語音的相對應時間。並藉由語音分析的音高追蹤技術與重音節計算，老師可以透過批改工具來審閱學生的發音分析，逐一字句地標示出學生發音上的各種錯誤：連音、語調、重音節、以及單字發音，幫助學生從視覺化的註解中，重新糾正發音上的錯誤，來達到自然發音訓練上的效果。

關鍵字

語音辨識；語音文字比對；發音批改

並列摘要

Chinese regard the full-scale learning and ability training more and more important since English e-Learning has been growing up in recent year. From early TOEFL test to nowadays GEPT and TOEIC examination, they have already included speaking to integrate English with the practicability and communication in our life. Now, training speaking skill becomes a major objective in English learning issue. With different reasoning on pronouncing system to native speakers, we usually speak the incorrect pronunciations that have caused some misunderstanding and difficulty of the communication when we talk to them. Our thesis applies some related techniques of speech recognition and string alignment which are integrated to our web-based spoken English training system. First the speech recognition engine analyzes each word to get its time code. And then the alignment between our speech and text is deduced through Speech-Text Alignment with Dynamic Programming. We also employ the speech analysis of pitch tracking and syllable stress calculation for computing pitches. With these procedures, the audio timestamp and pitch of each word is obtained. Teachers can check and approve the analysis of student’s speech on this correction system and mark a variety of speaking mistakes on each word: syllable stress, pronunciation, liaison and intonation. We purpose a system that can help students correct their speaking and pronouncing on English through visualized annotations and it also has good effects on training native spoken English.