簡易檢索 / 詳目顯示

回結果列表
研究生： 郭家成
GUO, JIA-CHENG
論文名稱： 基於梅爾倒頻譜特徵之智慧語音分類與長期語音一致性分析
Intelligent Voice Classification and Long-Term Voice Consistency Analysis Based on Mel-Frequency Cepstral Coefficients
指導教授： 舒宇宸
Shu, Yu-Chen
孫苑庭
SUN, YUAN-TING
學位類別： 碩士
Master
系所名稱： 理學院 - 數學系應用數學碩博士班
Department of Mathematics
論文出版年： 2026
畢業學年度： 114
語文別： 中文
論文頁數： 46
中文關鍵詞： 重症肌無力 、梅爾倒頻譜係數 、神經肌肉疾病 、支援向量機 、語音一致性 、聲音異常辨識 、人工智慧
外文關鍵詞： Myasthenia gravis, MFCC, Support vector machine, Voice consistency, Vocal fatigability, Artificial intelligence
相關次數： 點閱：11  下載：0 
分享至:
查詢本校圖書館目錄 查詢臺灣博碩士論文知識加值系統 勘誤回報

    • 重症肌無力是一種慢性自體免疫性神經肌肉疾病，患者常因口咽與喉部肌肉群的無力，表現出說話含糊及發聲疲乏等症狀。傳統臨床上對於嗓音異常的評估多仰賴醫師的主觀聽感，缺乏客觀且標準化的檢測工具。為此，本研究提出一套基於梅爾倒頻譜係數特徵之智慧語音分類與長期語音一致性分析框架，旨在提供非侵入式的 MG 早期輔助診斷與病情監測方法。

    在資料處理上，本研究採用「單一數字獨立切分」之技術，將受試者連續唸出的數字 1 至 10 獨立切分為 1 秒的音訊片段，不僅保留了完整的發音語意結構，更賦予了模型在特定數字上的臨床診斷解釋力。本研究主要分為兩大實驗：

    第一部分為使用支援向量機建立二元分類模型，再透過獨立兩樣本 t 檢定篩選出具統計顯著差異之數字（數字 4、8、10）。以新收案之未見資料進行驗證，實驗結果顯示，在僅保留具顯著區辨力之數字後，模型準確率由 73.33% 提升至 90.00%，F1-Score 自 0.5000 提升至 0.7692，強化了模型對未見資料的泛化能力。

    第二部分為結合重症肌無力日常生活活動量表中的延髓分數，將受試者依疾病嚴重度進行分層，並提出基於餘弦相似度的時序穩定度指標，以量化連續發聲過程中的衰退趨勢。使用單因子變異數分析與事後檢定結果證實，特定數字（數字 1、2、8）的時序特徵會隨患者病情嚴重度增加而產生差異。

    綜合兩項實驗結果，本研究發現「數字 8」在分類區辨與嚴重度追蹤上皆展現了高度的統計顯著性。本研究所建構之分析框架不僅能有效區別 MG 患者與健康對照組，其客觀的量化指標亦具備追蹤病情惡化趨勢之潛力，未來可望應用於 MG 患者的日常非侵入式自我檢測中。

    This study develops an objective, non-invasive framework for voice-based assessment of Myasthenia Gravis (MG) using Mel-Frequency Cepstral Coefficients (MFCC). Two complementary analytical models are proposed. Model I is a binary Support Vector Machine (SVM) classifier that distinguishes MG patients from healthy controls; using statistically screened digits (4, 8, and 10), it achieves 90.00% accuracy and an F1-score of 0.7692 on unseen data. Model II introduces a cosine-similarity-based temporal stability index to quantify vocal fatigability, revealing that specific digits (1, 2, and 8) show statistically significant differences across MG severity groups stratified by MG-ADL bulbar scores.Digit 8 consistently emerges as a discriminative indicator in both experiments, suggesting its potential as a reliable, non-invasive biomarker for MG monitoring and early detection.

    摘要 i Extended Abstract ii 誌謝 v Contents vi List of Tables ix List of Figures x 1 Introduction 1 1.1 Myasthenia Gravis, MG 1 1.1.1 Myasthenia Gravis Activities of Daily Living（MG-ADL） 1 1.2 Research Motivation 3 2 Preliminaries 4 2.1 Data Preprocessing 4 2.2 Feature Extraction 5 2.2.1 Pre-emphasis 5 2.2.2 Framing 6 2.2.3 Windowing 7 2.2.4 Discrete Fourier Transform, DFT 8 2.2.5 Log Energy and Discrete Cosine Transform 10 2.3 Support Vector Machine, SVM 11 2.3.1 Maximum Margin Hyperplane 11 2.4 Kernel Function 13 2.4.1 RBF Kernel 14 2.5 Independent Two-Sample t-test 15 2.5.1 Hypothesis Testing 15 2.5.2 Test Statistic and Pooled Variance 16 2.5.3 Significance Level and P-value 16 3 Methodology 17 3.1 Model I: MG Voice Classification using Support Vector Machine 17 3.1.1 Training Pipeline 17 3.1.2 Data Collection and Preprocessing 17 3.1.3 Feature Extraction 18 3.1.4 Model Training and Prediction 18 3.1.5 Statistical Identification of Sensitive Digits 19 3.2 Model II: Voice Consistency Analysis 19 3.2.1 Overview of Analytical Pipeline 19 3.2.2 Patient Categorization based on ADL Bulbar Score 20 3.2.3 Voice Consistency Modeling via Cosine Similarity 20 3.2.4 Statistical Evaluation and Group Analysis 21 4 Experimental Results 23 4.1 40 人 Model I 訓練結果與測試 23 4.2 基於 t 檢定之 Model I 對數字的區辨力 24 4.2.1 數字篩選規則 25 4.3 新收案資料之 Model I 驗證 26 4.3.1 篩選前表現（使用全部 10 個數字） 26 4.3.2 篩選後表現（保留數字 4、8、10） 27 4.3.3 套用篩選前後比較 27 4.4 Model II Testing 28 4.4.1 單因子變異數分析（One-way ANOVA） 28 4.4.2 Tukey Honestly Significant Difference 29 4.5 以分類機率驗證語音一致性指標之組間差異 30 5 Conclusion 32 5.1 Conclusion 32 Bibliography 33

    [1] Alexei Baevski, Henry Zhou, Abdelrahman Mohamed, and Michael Auli. wav2vec 2.0: A framework for self-supervised learning of speech representations. In Advances in Neural Information Processing Systems (NeurIPS), volume 33, pages 12449–12460, 2020.
    [2] B. E. Boser, I. M. Guyon, and V. N. Vapnik. A training algorithm for optimal margin classifiers. Proceedings of the Fifth Annual Workshop on Computational Learning Theory (COLT), 1992.
    [3] C. Cortes and V. Vapnik. Support-vector networks. Machine Learning, 20(3):273–297, 1995.
    [4] S. Davis and P. Mermelstein. Comparison of parametric representations for monosyllabic word recognition in continuously spoken sentences. IEEE Transactions on Acoustics, Speech, and Signal Processing, 28(4):357–366, 1980.
    [5] D. B. Drachman. Myasthenia gravis. New England Journal of Medicine, 330(25):1797–1810, 1994.
    [6] L. Dresser, R. Wlodarski, K. Rezania, et al. Myasthenia gravis: epidemiology, pathophysiology and clinical manifestations. Journal of Clinical Medicine, 10:2235, 2021.
    [7] A. Ganapathiraju, J. Hamaker, and J. Picone. Support vector machines for speech recognition. International Conference on Spoken Language Processing (ICSLP), 1998.
    [8] K. C. Golnik, R. Pena, A. G. Lee, et al. An ice test for the diagnosis of myasthenia gravis. Ophthalmology, 106:1282–1286, 1999.
    [9] F. Gomez, A. Mehra, E. Ensrud, D. Diedrich, and K. Laudanski. Covid-19: a modern trigger for guillain-barre syndrome, myasthenia gravis, and small fiber neuropathy. Frontiers in Neuroscience, 17:1198327, 2023.
    [10] A. Jayam Trouth, A. Dabi, N. Solieman, M. Kurukumbi, and J. Kalyanam. Myasthenia gravis: a review. Autoimmune Diseases, 2012:874680, 2012.
    [11] Mayo Clinic. Myasthenia gravis - symptoms and causes, n.d.
    [12] S. J. Oh, D. E. Kim, R. Kuruoglu, et al. Diagnostic sensitivity of the laboratory tests in myasthenia gravis. Muscle & Nerve, 15(6):720–724, 1992.
    [13] A. Punga, P. Maddison, J. Heckmann, et al. Epidemiology, diagnostics, and biomarkers of autoimmune neuromuscular junction disorders. The Lancet Neurology, 21(2):176–188, 2022.
    [14] C. Schneider-Gold and N. E. Gilhus. Advances and challenges in the treatment of myasthenia gravis. Therapeutic Advances in Neurological Disorders, 14:17562864211065406, 2021.
    [15] K. D. Sethi, M. H. Rivner, and T. R. Swift. Ice pack test for myasthenia gravis. Neurology, 37(8):1383–1385, 1987.
    [16] S. S. Stevens, J. Volkmann, and E. B. Newman. A scale for the measurement of the psychological magnitude pitch. The Journal of the Acoustical Society of America, 8(3):185–190, 1937.
    [17] B. R. Thanvi and T. C. N. Lo. Update on myasthenia gravis. Postgraduate Medical Journal, 80(950):690–700, 2004.
    [18] Jenthe Thienpondt, Caroline M. Speksnijder, and Kris Demuynck. Behavioral analysis of pathological speaker embeddings of patients during oncological treatment of oral cancer. In Proc. Interspeech 2023, pages 3018–3022, 2023.
    [19] A. Zanghì, P. S. Di Filippo, C. Rutigliano, C. Avolio, and E. D'Amico. Making the invisible audible: a real-world connected speech study in myasthenia gravis. Frontiers in Neurology, 17:1756913, 2026.

    QR CODE