臺灣博碩士論文加值系統

摘要
應用音節群組之加權型核心特徵矩陣於語音辨識替代錯誤修正
林立成* 吳宗憲**
國立成功大學資訊工程學系

近年來，語音辨識技術的成熟使其在科技領域的應用上蓬勃發展。然而，由於語音辨識技術在複雜環境等不同變數的影響之下，使其辨識率依然受到限制。所造成的語音辨識的錯誤輸出因此影響了語音辨識技術的實用價值。
因此，本論文透過後處理的方式，利用自然語言處理方式提出前後文相關之音節群組（Context Dependent Syllable Cluster、簡稱CDSC）為主的音節群組之加權型核心特徵矩陣，透過所提出此前後文相關之音節利用發音屬性進行分群，計算核心特徵矩陣結合音節混淆矩陣所得出的音節群組之加權型核心特徵矩陣，來找出最為相似性之音節，以用來改善語音辨識器輸出結果之辨識率。
利用本論文所提出之音節群組之加權型核心特徵矩陣，對語音辨識器之錯誤輸出結果進行修正，能分別將詞錯誤率由原本48.50%降低至45.31%及音節錯誤率由原本15.37%降低至10.31%。實驗結果說明了本論文所提出之方法，能有效地對語音辨識器所輸出的替代錯誤進行修正。


關鍵字－前後文相關之音節，語音辨識，錯誤修正，自然語言處理
*作者 **指導教授

Abstract
A Syllable Cluster Based Weighted Kernel Feature Matrix for ASR Substitution Error Correction
David Sarwono * Chung-Hsien Wu**
Institute of Computer Science and Information Engineering,
National Cheng Kung University, Tainan, Taiwan, R.O.C.

In recent years Automatic Speech Recognition (ASR) technology has become one of the most growing technologies in engineering science and research. However, the performance of ASR technology is still restricted in adverse environments. Errors in Automatic Speech Recognition outputs lead to low performance for speech applications, therefore correction techniques for these errors will be beneficial to applications relied on ASR outputs.
In this study, A Syllable Cluster Based Weighted Kernel Feature Matrix based on Context Dependent Syllable Cluster (CDSC) is proposed for the generation of correction candidates. For candidate selection in the second stage, the n-gram language model is used to determine the final corrected sentence output, thus to improve speech recognition output results recognition rate.
Experiments show that the proposed method improved from 48.50% to 45.31% and 15.37% to 10.31% in terms of Word Error Rate score and Syllable Error Rate as compared to the speech recognition approach.


Keyword－Context Dependent Syllable, Automatic Speech Recognizer, Error Correction, Natural Language Processing
* The Author ** The Advisor

目錄
致謝 I
摘要 II
Abstract III
目錄 IV
圖目錄 VII
表目錄 VIII
第一章 緒論 1
1.1 研究背景 1
1.2 研究動機 2
1.3 研究目的 4
1.4 論文架構 5
第二章 文獻回顧 7
第三章 系統架構 10
3.1 語音辨識替代錯誤修正模型 10
3.2 語音辨識替代錯誤修正系統架構 13
第四章 研究方法 15
4.1 使用語料 16
4.2 前後文相關音節 18
4.3 發音屬性 19
4.4 語音辨識音節擴展 22
4.4.1語音辨識 22
4.4.2訓練流程 22
4.5 音節群組之加權型核心特徵矩陣 24
4.5.1訓練語音檔案切割 26
4.5.2最大熵為基礎之決策樹分群 27
4.5.3辨識信心分數 30
4.5.4前後文相關核心特徵混淆分數 31
4.5.5建立音節混淆矩陣 32
4.5.6建立音節群組之加權型核心特徵矩陣 33
4.6 最佳路徑及分數評估 36
4.6.1最佳路徑 36
4.6.2音節文字轉換 37
4.6.3分數評估 38
第五章 實驗評估 40
5.1 實驗設計 40
5.2 實驗結果與分析 43
5.3 比較系統實驗設定及實驗結果 46
第六章 結論及未來展望 48
6.1 結論 48
6.2 未來研究 48
參考文獻 50
作者簡介 53

參考文獻
[1]Dufour R., Estève Y., Correcting ASR outputs: specific solutions to specific errors in French, in IEEE Workshop on Spoken Language Technology (SLT 2008), Goa, India, 2008, 15-18.

[2]Sagawa, H., Mitamura, T., and Nyberg, E., Correction grammars for error handling in a speech dialog system, in Proceedings of HLT-NAACL 2004: Short Papers, Boston, 2004, 61-64.

[3]López-Cózar, R. and Callejas, Z., ASR post-correction for spoken dialogue systems based on semantic, syntactic, lexical and contextual information, Speech Communication, 50(8-9): 745-766, 2008.

[4]Brockett, C., Dolan, W. B., and Gamon, M., Correcting ESL Errors using Phrasal SMT Techniques, in Proceedings of the 1st International Conference on Computational Linguistics and the 44th annual meeting of the ACL, 2006, 249-256.

[5]Liu, C.-H. and Wu, C.-H., Prosodic word-based error correction in speech recognition using prosodic word expansion and contextual information, in Proceedings of INTERSPEECH-2010, Makuhari, Chiba, Japan, 2010, 1385-1388.

[6]Wu, C.-H., Liu, C.-H., Harris, M., and Yu, L.-C., Sentence Correction Incorporating Relative Position and Parse Template Language Models, IEEE Trans. on Audio, Speech and Language Processing, 18(6): 1170-1181, 2010.

[7]Weidong Zhou., Baozong Yuan., Zhenjiang Miao., Xiaofang Tang., Error correction via phonetic similarity-based processing for chinese spoken dialogue system, Signal Processing, 2006 8th International Conference on , vol.3, no., 16-20 2006.

[8]MATBN中文廣播新聞語料庫簡介
[9]Chi-Tien Chiu, Jia-Jang Tu, Jeng-Shien Lin, Sen-Chia Chang., Phone-based Mandarin Speech Recognition, ICL TECHNICAL JOURNAL,2007, 37-41.

[10]謝國平，“語言學概論，三民出版社，民國87年

[11]蔡沛任，應用語音屬性分析於構音障礙者之發音錯誤與修正回饋，碩士論文，國立成功大學資訊工程系，2007

[12]International Phonetic Association (IPA), Handbook.

[13]Wikipedia-Viterbi algorithm
http://en.wikipedia.org/wiki/Viterbi_algorithm

[14]游政仁，語者分群及語音命令辨識之研究，碩士論文，國立台灣科，大學資訊工程系，2008
[15]溫家誠，多媒體應用之與音辨識系統，碩士論文，國立中央大學電機工程研究所，2008
[16]Jie Wang, Haiping Lu, K. N. Plataniotis, Juwei Lu, Gaussian Kernel Optimization for Pattern Classification, Elsevier Science, 2008.

[17]Elena Deza, Michel Marie Deza,Encyclopedia of Distances, 2009, page 94, Springer.

[18] Sha Meng, Peng Yu, Frank Seide, and Jia Liu, A study of lattice-based spoken term detection for chinese spontaneous speech, in ASRU, 2007.

[19]Wikipedia-Dynamic Programming
http://en.wikipedia.org/wiki/Dynamic_programming

[20]Iain McCowan, Darren Moore, John Dines, Daniel Gatica-Perez, Mike Flynn, Pierre Wellner, Herv_e Bourlard, On the Use of Information Retrieval Measures forSpeech Recognition Evaluation, 2005.

[21]Xinhui Li, Xiangdong Wang, Yueliang Qian, Shouxun Lin, Candidate generation for interactive Chinese speech recognition, Pervasive Computing (JCPC), 2009 Joint Conferences on , vol., no., pp.583-588, 3-5 Dec. 2009.