第五章 結論與未來展望
5.2 未來展望
語料的選擇上，外文語言資料庫選擇較為豐富，相對的中文語言資料庫選擇較少，而本 論文使用陳小娟老師余 2002 年提出 MSPIN(Mandarin Speech Perception In Noise)中文語言資 料庫，總共 300 句語句，分為高預測性語句與低預測語句，實驗方式為辨識語句中的最後一 個字。由於雜訊消除為整段語句，而我們使用此語料庫只需辨別最後一個字是否正確，雖然 在高預測性語句中含有 2~3 個關鍵字，並允許受測者對語句推導猜測，但低預測性語句並沒 有關鍵字可供受測者推導猜測，這樣的測試方式，並不能完全為語句辨識，而有單字詞辨識 的情境產生，而我們所提出的噪音抑制法使用此中文語料庫，辨識率皆有所成效，但是在之 後可以選擇其他的中文語料庫，以測試方法是否依然有效提升辨識率。我們目前也有針對其 他中文語料庫，進行錄製與嚴謹的調整測試，例如：TMHINT(Taiwan Mandarin Hearing in Noise Test)，由黃銘緯於 2005 年依據 HINT 的實驗環境所建立的中文語料庫[31]，但 TMHINT 實驗 方式為整段語句辨識，所以需要撰寫軟體介面以方便進行實驗設置及操作，目前也正在著手 實作。
本篇的噪音抑制方法以雜訊消除和保留語音完整性為出發點，實驗結果語音辨識力有所 提升，經由 rANOVA 分析後，我們所提出的方法與 Original 有顯著差異，但我們所提出的方 法與其他的方法只有少部分有顯著差異，也就是說我們所提出的方法並沒有明顯的優勢，所 以未來如何改進我們的方法，抑或是發展出更適用於人工電子耳使用者的噪音抑制方法，以 提高人工電子耳使用者的語音辨識能力。
43
頻 域 獨 立 成 份 分 析 法 在 使 用 上 仍 然 存 在 運 算 時 間 較 長 的 問 題 ， 即 使 我 們 使 用 complex-valued FastICA 來縮短運算時間，並且將分頻方式依照人耳耳蝸頻率分布圖減少不必 要的頻帶計算，結果顯示聲音訊號經由此方法將雜訊消除，且保留語音完整性，提升辨識率，
但是運算時間相較於時域獨立成份分析法，依然需要較長的時間，因此如何分頻以及頻帶的 選擇上，還有許多選擇及改進的空間。
受測者的選擇，在本篇論文皆以正常聽力者進行實驗，受測者所聽到的聲音皆由 HRTFs 模擬具有方向性的雙耳聲音，再經過 vocoder 進行編碼，雖然在編碼策略已經盡可能的模擬 真實的人工電子耳使用者所聽到的聲音，但是有許多論文顯示，許多方法用於實際的人工電 子耳使用者後，辨識率皆比模擬訊號辨識率低，因此能夠找到人工電子耳使用者成為我們的 受測者，更能夠調整我們的噪音抑制法，來提升辨識率。
44
參考文獻
[1] P. C. Loizou, “Mimicking the human ear,” IEEE Signal Processing Magazine, vol. 15, issue 5, pp. 101-103, 1998.
[2] P. C. Loizou, “Signal processing techniques for cochlear implants,” IEEE Engineering in Medicine and Biology Magazine, vol. 18, no. 3, pp. 34-46, 1999.
[3] B.S. Wilson and M.F. Dorman, “Cochlear implants: current designs and future possibilities,”
Journal of Rehabilitation Research & Development, vol. 45, no. 5, pp. 695-730, Feb. 2008.
[4] B. C. Moore, Cochlear hearing loss. Whurr, 1998.
[5] B. Wilson, D. Lawson, and M. Zerbi, “Advances in coding strategies for cochlear implants,”
Advances in Otolaryngology - Head and Neck Surgery, vol. 9, pp. 105-129, 1995.
[6] H. Hakin and B.V. Veen, Signals and Systems, 2nd edition, John Wiley & Sons. Inc. 2003.
[7] S. G. Mallat, “A theory for multiresolution signal decomposition: The wave representation,”
Communications on Pure and Applied Mathematics , vol. 41, pp. 674-693,1988.
[8] 董英凱，「基於小波轉換之語音增強系統」，國立成功大學電機工程學系碩博士班論文，
2005.
[9] N. Wiener, “Extrapolation interpolation, and smoothing of stationary time series with engineering applications,” Cambridge, MA: MIT Press, 1949.
[10] A. Grossmann and J. Morlet, “Decomposition of hardy functions into square integrable wavelets of constant shape,” SIAM J Math Analysis, vol. 15, no. 4, pp. 723–736, 1984.
[11] Y. Hu and P. C. Loizou, “Speech enhancement based on wavelet thresholding the multitaper spectrum,” IEEE Transaction on Speech Audio Process, vol. 12, no. 1, pp. 59-67, 2004.
[12] A. Maravall, “Minimum mean squared error estimation of the noise in unobserved component models,” Journal of Business and Economic Statistics, vol. 5, no. 1, pp. 115-120, 1987.
45
[13] A. Hyvärinen, J. Karhunen and E. Oja, “Independent Component Analysis,” Wiley InterScience, 2001.
[14] H. Saruwatari, T. Kawamura, K. Sawai, A. Kaminuma and M. Sakata, “Blind source separation based on fast-convergence algorithm using ICA and beamforming for real convolutive mixture,”
IEEE International Conference on Acoustics, Speech, and Signal Processing (ICASSP), vol. 1, pp. 921-924, 2002.
[15] T. Nishikawa, H. Saruwatari and K. Shikano, “Bund source separation based on Multi-Stage ICA combining frequency-domain ICA and time-domain ICA,” IEEE International Conference on Acoustics, Speech, and Signal Processing (ICASSP), vol. 1, pp.917-920, 2002.
[16] 陳彥名，「以獨立成份分析法萃取背景噪音中語音訊號之研究—於助聽器可能之應用」，
國立陽明醫學大學醫學工程研究所碩士論文，2004。
[17] 連憶如，「頻域獨立成份分析法於語音訊號分離之研究」，國立交通大學電機與控制工程 研究所論文，2004。
[18] A.J. Bell and T.J. Sejnowski, “A non-linear information maximization algorithm that performs blind separation.” In Advances in Neural Information Processing System 7, pp. 467-474, The MIT Press, Cambridge, MA, 1995.
[19] A.J. Bell and T.J. Sejnowski, “An information-maximization approach to blind separation and blind deconvolution.” Neural Computation, vol. 7, issue. 6, pp. 1129-1159, 1995.
[20] A. Hyvärinen, “A family of fixed-point algorithms for independent component analysis.” IEEE International Conference on Acoustics, Speech, and Signal Processing (ICASSP), vol. 5, pp.
3917-3920, 1997.
[21] A. Hyvärinen and E. Oja. “A fast fixed-point algorithm for independent component analysis,”
Neural Computation, vol. 9, no. 7, pp. 1483-1492, 1997.
[22] A. Hyvärinen, “Survey on independent component analysis,” Neural Computing Surveys, vol.
46
2, pp. 94-128, 1999.
[23] A. Hyvärinen, “Fast and robust fixed-point algorithms for independent component analysis,”
IEEE Transactions on Neural Networks, vol. 10, no. 3, pp. 626-634, 1999.
[24] E. Bingham and A. Hyvärinen, “A fast fixed-point algorithm for independent component analysis of complex valued signals,” International Journal of Neural Systems, vol. 10, no. 1, pp.
1-8, Feb. 2000.
[25] H. Sawada, R. Mukai, S. Araki and S. Makino, “A robust and precise method for solving the permutation problem of frequency-domain blind source separation,” IEEE Transaction Speech and Audio Processing, vol.12, no. 5, pp. 530-538, Sep. 2004
[26] 葉旭輝，「華語各頻帶訊息對語音理解之重要度分析」，國立陽明大學醫學工程研究所論 文，2005。
[27] E. Zwicker and E. Terhardt, “Analytical expressions for critical-band rate and critical bandwidth as a function of frequency,” Institute of Electro Acoustics, vol. 68, issue 5, pp.
1523-1525, 1980.
[28] C. T. M. Choi , C. H. Hsu, W. Y. Tsai and Y. H. Lee , “A Vocoder for a Novel Cochlear Implant Stimulating Strategy based on Virtual Channel Technology, ” Proceedings of 13th International Conference on Biomedical Engineering, pp. 310-313, 2008.
[29] 蔡志浩和陳小娟，「噪音背景辨識語音測驗編製研究」，特殊教育研究學刊，第 23 卷，第 121-140 頁，2002。
[30] B. Gardner and K. Martin, “HRTF measurements of a KEMAR dummy-head microphone,”
MIT Media Lab Perceptual Computing, Technical Report #280, May, 1994.
[31] 黃銘緯，「台灣地區噪音下漢語語音聽辨測試」，國立台北護理學院聽語障礙科學研究所 論文，2005.
47
附錄 A 聽覺測試中文語料列表
實驗使用 MSPIN(Mandarin Speech Perception In Noise)中文語料表單，由陳小娟老師於 2002 年發表[29]。
48
49
50
51
52
53
附錄 B 第一階段實驗詳細數據
受測者一語音辨識正確率(%)
Noise：Multi-talker babble Speech：High predictability
S000N270 S045N315
SNR-5 SNR0 SNR5 SNR-5 SNR0 SNR5
Original 0 8.33 83.33 0 0 16.67
TD_ICA 16.67 50 58.33 0 50 75
FD_DFT 33.33 58.33 100 33.33 41.67 91.67 FD_CTB 8.33 58.33 75 25 41.67 100
Noise：Multi-talker babble Speech：Low predictability
S000N270 S045N315
SNR-5 SNR0 SNR5 SNR-5 SNR0 SNR5
Noise：Multi-talker babble Speech：High predictability
S000N270 S045N315
SNR-5 SNR0 SNR5 SNR-5 SNR0 SNR5
Original 0 41.67 75 0 0 25
TD_ICA 8.33 41.67 41.67 16.67 75 58.33 FD_DFT 41.67 83.33 91.67 16.67 75 100
FD_CTB 25 75 100 58.33 91.67 75
Noise：Multi-talker babble Speech：Low predictability
S000N270 S045N315
SNR-5 SNR0 SNR5 SNR-5 SNR0 SNR5
Original 0 8.33 33.33 0 0 25
TD_ICA 8.33 58.33 50 0 16.67 58.33
FD_DFT 25 25 66.67 33.33 25 50
FD_CTB 58.33 66.67 58.33 33.33 66.67 91.67
54
受測者三語音辨識正確率(%)
Noise：Multi-talker babble Speech：High predictability
S000N270 S045N315
SNR-5 SNR0 SNR5 SNR-5 SNR0 SNR5 Original 0 16.67 66.67 8.33 0 16.67 TD_ICA 8.33 50 75 16.67 41.67 66.67 FD_DFT 33.33 75 83.33 25 83.33 66.67 FD_CTB 25 83.33 83.33 8.33 75 83.33
Noise：Multi-talker babble Speech：Low predictability
S000N270 S045N315
SNR-5 SNR0 SNR5 SNR-5 SNR0 SNR5
Noise：Multi-talker babble Speech：High predictability
S000N270 S045N315
SNR-5 SNR0 SNR5 SNR-5 SNR0 SNR5
Original 0 8.33 50 0 0 8.33
TD_ICA 8.33 25 58.33 16.67 33.33 75 FD_DFT 8.33 83.33 75 16.67 58.33 83.33
FD_CTB 25 75 66.67 8.33 75 75
Noise：Multi-talker babble Speech：Low predictability
S000N270 S045N315
SNR-5 SNR0 SNR5 SNR-5 SNR0 SNR5
Original 0 8.33 41.67 0 0 0
TD_ICA 0 16.67 41.67 8.33 25 58.33 FD_DFT 16.67 33.33 66.67 25 33.33 41.67 FD_CTB 25 58.33 66.67 8.33 41.67 75
55
受測者五語音辨識正確率(%)
Noise：Multi-talker babble Speech：High predictability
S000N270 S045N315
SNR-5 SNR0 SNR5 SNR-5 SNR0 SNR5
Original 0 8.33 58.33 0 0 8.33
TD_ICA 8.33 33.33 83.33 0 41.67 100 FD_DFT 41.67 83.33 75 25 50 91.67 FD_CTB 25 83.33 83.33 41.67 75 91.67
Noise：Multi-talker babble Speech：Low predictability
S000N270 S045N315
SNR-5 SNR0 SNR5 SNR-5 SNR0 SNR5
Noise：Multi-talker babble Speech：High predictability
S000N270 S045N315
SNR-5 SNR0 SNR5 SNR-5 SNR0 SNR5
Original 0 8.33 58.33 0 0 16.67
TD_ICA 8.33 41.67 41.67 8.33 25 66.67 FD_DFT 16.67 50 75 8.33 66.67 100 FD_CTB 8.33 75 91.67 58.33 75 91.67
Noise：Multi-talker babble Speech：Low predictability
S000N270 S045N315
SNR-5 SNR0 SNR5 SNR-5 SNR0 SNR5
Original 0 25 50 0 0 0
TD_ICA 8.33 50 58.33 16.67 25 50 FD_DFT 16.67 41.67 41.67 16.67 8.33 50 FD_CTB 8.33 66.67 75 33.33 58.33 75
56
受測者七語音辨識正確率(%)
Noise：Multi-talker babble Speech：High predictability
S000N270 S045N315
SNR-5 SNR0 SNR5 SNR-5 SNR0 SNR5
Original 0 16.67 66.67 0 0 8.33
TD_ICA 8.33 41.67 58.33 8.33 50 41.67 FD_DFT 8.33 75 83.33 33.33 66.67 91.67 FD_CTB 25 66.67 83.33 16.67 58.33 100
Noise：Multi-talker babble Speech：Low predictability
S000N270 S045N315
SNR-5 SNR0 SNR5 SNR-5 SNR0 SNR5
Noise：Multi-talker babble Speech：High predictability
S000N270 S045N315
SNR-5 SNR0 SNR5 SNR-5 SNR0 SNR5
Original 0 8.33 58.33 0 0 25
TD_ICA 8.33 58.33 58.33 8.33 50 66.67 FD_DFT 41.67 66.67 83.33 33.33 75 91.67 FD_CTB 41.67 83.33 83.33 25 33.33 75
Noise：Multi-talker babble Speech：Low predictability
S000N270 S045N315
SNR-5 SNR0 SNR5 SNR-5 SNR0 SNR5
Original 0 25 58.33 0 0 0
TD_ICA 8.33 25 58.33 8.33 25 50
FD_DFT 16.67 75 50 0 33.33 41.67
FD_CTB 8.33 50 66.67 50 41.67 50
57
受測者九語音辨識正確率(%)
Noise：Multi-talker babble Speech：High predictability
S000N270 S045N315
SNR-5 SNR0 SNR5 SNR-5 SNR0 SNR5
Original 0 16.67 33.33 0 0 0
TD_ICA 8.33 25 58.33 0 16.67 41.67 FD_DFT 0 66.67 66.67 25 41.67 66.67
FD_CTB 33.33 50 66.67 0 50 83.33
Noise：Multi-talker babble Speech：Low predictability
S000N270 S045N315
SNR-5 SNR0 SNR5 SNR-5 SNR0 SNR5
Noise：Multi-talker babble Speech：High predictability
S000N270 S045N315
SNR-5 SNR0 SNR5 SNR-5 SNR0 SNR5
Original 0 8.33 91.67 0 0 16.67
TD_ICA 16.67 33.33 83.33 16.67 58.33 91.67
FD_DFT 25 75 91.67 25 66.67 100
FD_CTB 33.33 66.67 83.33 41.67 66.67 75 Noise：Multi-talker babble Speech：Low predictability
S000N270 S045N315
SNR-5 SNR0 SNR5 SNR-5 SNR0 SNR5 Original 8.33 33.33 50 0 0 16.67
TD_ICA 8.33 25 41.67 8.33 25 50
FD_DFT 25 50 83.33 33.33 41.67 50 FD_CTB 41.67 75 66.67 8.33 66.67 41.67
58
附錄 C 第二階段實驗詳細數據
Wiener filter 參數組合列表
Condition NoiseMargin Hangover Alpha SNRdiff
C001 1 2 0.5 0.5
Condition Monther cf
C001 db9 0.1
C002 db9 0.7
C003 db9 1.2
C004 db9 1.6
59
C001 C002 C003 C004 C005 C006 C007 C008 C009 C010 C011 C012 C013 C014 C015 C016 C017 C018 C019 C020 C021 C022 C023 C024 Ori
Wiener
C001 C002 C003 C004 C005 C006 C007 C008 C009 C010 C011 C012 C013 C014 C015 C016 C017 C018 C019 C020 C021 C022 C023 C024 Ori
Wiener
60
C001 C002 C003 C004 Ori
Wavlet
C001 C002 C003 C004 Ori
Wavlet
61
附錄 D 第三階段實驗詳細數據
受測者一語音辨識正確率(%)
Noise：Multi-talker babble Speech：High predictability
S000N270 S000N030
SNR0 SNR5 SNR10 SNR-5 SNR0 SNR5 Original 7.14 50 85.71 7.14 57.14 78.57 win 14.29 64.29 92.86 0 50 85.71 FD_win 28.57 64.29 92.86 21.43 64.29 78.57 wlt 14.29 57.14 71.43 21.43 42.86 78.57 FD_wlt 14.29 57.14 85.71 21.43 50 78.57
Noise：Multi-talker babble Speech：Low predictability
S000N270 S000N030
SNR0 SNR5 SNR10 SNR-5 SNR0 SNR5 Original 7.14 28.57 64.29 14.29 21.43 42.86 win 14.29 21.43 71.43 21.43 28.57 57.14 FD_win 14.29 28.57 57.14 7.14 21.43 50
wlt 14.29 50 50 14.29 28.57 42.86 FD_wlt 7.14 42.86 71.43 14.29 28.57 42.86
62
受測者二語音辨識正確率(%)
Noise：Multi-talker babble Speech：High predictability
S000N270 S000N030
SNR0 SNR5 SNR10 SNR-5 SNR0 SNR5 Original 14.29 57.14 78.57 14.29 57.14 85.71 win 14.29 42.86 92.86 14.29 57.14 78.57 FD_win 14.29 57.14 71.43 21.43 50 71.43 wlt 7.14 64.29 85.71 14.29 57.14 71.43 FD_wlt 14.29 50 78.57 21.43 57.14 64.29
Noise：Multi-talker babble Speech：Low predictability
S000N270 S000N030
SNR0 SNR5 SNR10 SNR-5 SNR0 SNR5 Original 7.14 35.71 64.29 7.14 35.71 78.57 win 28.57 50 71.43 7.14 35.71 78.57 FD_win 21.43 50 71.43 14.29 35.71 71.43 wlt 7.14 42.86 42.86 7.14 35.71 50 FD_wlt 21.43 50 35.71 14.29 35.71 64.29
63
受測者三語音辨識正確率(%)
Noise：Multi-talker babble Speech：High predictability
S000N270 S000N030
SNR0 SNR5 SNR10 SNR-5 SNR0 SNR5 Original 7.14 50 71.43 7.14 28.57 85.71 win 21.43 71.43 78.57 35.71 57.14 100 FD_win 21.43 64.29 85.71 21.43 64.29 92.86
wlt 21.43 57.14 85.71 21.43 57.14 85.71 FD_wlt 28.57 64.29 78.57 21.43 42.86 85.71
Noise：Multi-talker babble Speech：Low predictability
S000N270 S000N030
SNR0 SNR5 SNR10 SNR-5 SNR0 SNR5 Original 7.14 57.14 42.86 21.43 35.71 57.14 win 7.14 64.29 78.57 21.43 50 64.29 FD_win 7.14 50 85.71 21.43 42.86 64.29 wlt 14.29 42.86 71.43 14.29 35.71 57.14 FD_wlt 14.29 50 78.57 14.29 21.43 50
64
受測者四語音辨識正確率(%)
Noise：Multi-talker babble Speech：High predictability
S000N270 S000N030
SNR0 SNR5 SNR10 SNR-5 SNR0 SNR5 Original 21.43 42.86 85.71 14.29 50 85.71 win 14.29 64.29 92.86 28.57 71.43 85.71 FD_win 21.43 42.86 92.86 28.57 71.43 100
wlt 21.43 42.86 71.43 28.57 71.43 78.57 FD_wlt 21.43 57.14 85.71 21.43 64.29 85.71
Noise：Multi-talker babble Speech：Low predictability
S000N270 S000N030
SNR0 SNR5 SNR10 SNR-5 SNR0 SNR5 Original 14.29 35.71 71.43 7.14 42.86 71.43
win 14.29 50 42.86 28.57 42.86 50 FD_win 7.14 42.86 64.29 28.57 35.71 78.57
wlt 21.43 28.57 50 14.29 42.86 57.14 FD_wlt 7.14 42.86 57.14 14.29 42.86 57.14
65
受測者五語音辨識正確率(%)
Noise：Multi-talker babble Speech：High predictability
S000N270 S000N030
SNR0 SNR5 SNR10 SNR-5 SNR0 SNR5 Original 0 50 85.71 14.29 57.14 71.43
win 0 50 100 14.29 64.29 85.71
FD_win 14.29 50 92.86 14.29 64.29 100 wlt 14.29 42.86 71.43 21.43 42.86 50 FD_wlt 14.29 42.86 85.71 28.57 57.14 57.14
Noise：Multi-talker babble Speech：Low predictability
S000N270 S000N030
SNR0 SNR5 SNR10 SNR-5 SNR0 SNR5 Original 7.14 57.14 64.29 14.29 21.43 42.86 win 7.14 35.71 71.43 14.29 28.57 71.43 FD_win 0 50 71.43 14.29 42.86 78.57 wlt 14.29 42.86 71.43 14.29 21.43 71.43 FD_wlt 14.29 50 71.43 14.29 35.71 71.43
66
受測者六語音辨識正確率(%)
Noise：Multi-talker babble Speech：High predictability
S000N270 S000N030
SNR0 SNR5 SNR10 SNR-5 SNR0 SNR5 Original 7.14 57.14 64.29 0 28.57 64.29 win 7.14 57.14 100 0 28.57 78.57 FD_win 7.14 50 71.43 7.14 35.71 92.86 wlt 7.14 50 64.29 7.14 21.43 28.57 FD_wlt 14.29 50 64.29 14.29 35.71 50
Noise：Multi-talker babble Speech：Low predictability
S000N270 S000N030
SNR0 SNR5 SNR10 SNR-5 SNR0 SNR5 Original 7.14 50 64.29 7.14 28.57 50
win 7.14 35.71 71.43 14.29 21.43 64.29
FD_win 7.14 35.71 50 0 35.71 50
wlt 0 35.71 42.86 7.14 21.43 35.71 FD_wlt 7.14 35.71 50 14.29 28.57 35.71
67
受測者七語音辨識正確率(%)
Noise：Multi-talker babble Speech：High predictability
S000N270 S000N030
SNR0 SNR5 SNR10 SNR-5 SNR0 SNR5 Original 0 57.14 71.43 28.57 64.29 64.29 win 7.14 42.86 71.43 28.57 64.29 85.71 FD_win 14.29 50 78.57 28.57 57.14 71.43 wlt 7.14 42.86 78.57 0 57.14 64.29 FD_wlt 7.14 57.14 64.29 0 42.86 64.29
Noise：Multi-talker babble Speech：Low predictability
S000N270 S000N030
SNR0 SNR5 SNR10 SNR-5 SNR0 SNR5 Original 7.14 21.43 50 7.14 28.57 57.14 win 7.14 35.71 57.14 14.29 42.86 64.29 FD_win 0 35.71 50 28.57 28.57 71.43 wlt 7.14 42.86 50 21.43 21.43 57.14 FD_wlt 7.14 35.71 57.14 21.43 14.29 78.57
68
受測者八語音辨識正確率(%)
Noise：Multi-talker babble Speech：High predictability
S000N270 S000N030
SNR0 SNR5 SNR10 SNR-5 SNR0 SNR5 Original 14.29 42.86 92.86 14.29 50 85.71 win 14.29 71.43 85.71 35.71 71.43 100 FD_win 14.29 64.29 92.86 28.57 71.43 85.71
wlt 14.29 57.14 71.43 14.29 57.14 100 FD_wlt 14.29 42.86 78.57 21.43 42.86 78.57
Noise：Multi-talker babble Speech：Low predictability
S000N270 S000N030
SNR0 SNR5 SNR10 SNR-5 SNR0 SNR5 Original 21.43 28.57 50 28.57 42.86 71.43 win 14.29 50 78.57 14.29 57.14 42.86 FD_win 35.71 42.86 78.57 0 57.14 78.57 wlt 0 42.86 71.43 21.43 21.43 71.43 FD_wlt 21.43 28.57 78.57 14.29 28.57 57.14
69
受測者九語音辨識正確率(%)
Noise：Multi-talker babble Speech：High predictability
S000N270 S000N030
SNR0 SNR5 SNR10 SNR-5 SNR0 SNR5 Original 21.43 42.86 92.86 0 21.43 64.29 win 14.29 50 100 7.14 28.57 71.43 FD_win 28.57 57.14 100 7.14 28.57 78.57 wlt 0 57.14 85.71 7.14 35.71 64.29 FD_wlt 21.43 57.14 85.71 7.14 50 71.43
Noise：Multi-talker babble Speech：Low predictability
S000N270 S000N030
SNR0 SNR5 SNR10 SNR-5 SNR0 SNR5 Original 14.29 42.86 71.43 0 14.29 50
win 7.14 35.71 64.29 0 28.57 50 FD_win 14.29 35.71 71.43 0 28.57 50 wlt 21.43 35.71 57.14 7.14 21.43 28.57 FD_wlt 14.29 35.71 64.29 14.29 28.57 42.86
70
受測者十語音辨識正確率(%)
Noise：Multi-talker babble Speech：High predictability
S000N270 S000N030
SNR0 SNR5 SNR10 SNR-5 SNR0 SNR5 Original 21.43 50 78.57 21.43 50 71.43 win 21.43 57.14 100 21.43 57.14 92.86 FD_win 21.43 50 100 21.43 50 78.57 wlt 14.29 42.86 78.57 14.29 50 64.29 FD_wlt 21.43 57.14 71.43 21.43 42.86 85.71
Noise：Multi-talker babble Speech：Low predictability
S000N270 S000N030
SNR0 SNR5 SNR10 SNR-5 SNR0 SNR5 Original 7.14 35.71 64.29 21.43 28.57 57.14 win 14.29 57.14 71.43 7.14 42.86 57.14 FD_win 14.29 57.14 71.43 28.57 42.86 64.29 wlt 7.14 42.86 50 14.29 35.71 64.29 FD_wlt 7.14 42.86 64.29 14.29 57.14 42.86
71
受測者十一語音辨識正確率(%)
Noise：Multi-talker babble Speech：High predictability
S000N270 S000N030
SNR0 SNR5 SNR10 SNR-5 SNR0 SNR5 Original 7.14 42.86 78.57 14.29 42.86 78.57 win 14.29 57.14 92.86 14.29 50 78.57 FD_win 14.29 57.14 92.86 14.29 57.14 92.86 wlt 7.14 57.14 78.57 14.29 57.14 71.43 FD_wlt 14.29 57.14 92.86 21.43 64.29 57.14
Noise：Multi-talker babble Speech：Low predictability
S000N270 S000N030
SNR0 SNR5 SNR10 SNR-5 SNR0 SNR5 Original 7.14 35.71 71.43 7.14 35.71 50
win 7.14 35.71 71.43 14.29 50 50 FD_win 14.29 42.86 71.43 7.14 57.14 57.14
wlt 0 42.86 71.43 14.29 50 35.71 FD_wlt 14.29 42.86 71.43 21.43 50 50
72
受測者十二語音辨識正確率(%)
Noise：Multi-talker babble Speech：High predictability
S000N270 S000N030
SNR0 SNR5 SNR10 SNR-5 SNR0 SNR5 Original 14.29 42.86 78.57 14.29 50 85.71 win 14.29 64.29 71.43 21.43 50 92.86 FD_win 14.29 64.29 92.86 28.57 50 92.86 wlt 21.43 50 78.57 14.29 42.86 64.29 FD_wlt 21.43 57.14 85.71 14.29 57.14 85.71
Noise：Multi-talker babble Speech：Low predictability
S000N270 S000N030
SNR0 SNR5 SNR10 SNR-5 SNR0 SNR5 Original 7.14 35.71 42.86 0 21.43 42.86 win 7.14 50 50 21.43 42.86 71.43 FD_win 7.14 57.14 64.29 21.43 42.86 71.43 wlt 14.29 35.71 42.86 7.14 28.57 42.86 FD_wlt 7.14 42.86 50 14.29 28.57 57.14