| { |
| "paper_id": "O10-1006", |
| "header": { |
| "generated_with": "S2ORC 1.0.0", |
| "date_generated": "2023-01-19T08:06:52.325298Z" |
| }, |
| "title": "A Study of Minimum Variance Modulation Filter for Robust Speech Recognition", |
| "authors": [ |
| { |
| "first": "Ren-Hau", |
| "middle": [], |
| "last": "\u8b1d\u4ec1\u8c6a", |
| "suffix": "", |
| "affiliation": {}, |
| "email": "" |
| }, |
| { |
| "first": "", |
| "middle": [], |
| "last": "Hsieh", |
| "suffix": "", |
| "affiliation": {}, |
| "email": "" |
| }, |
| { |
| "first": "Hao-Teng", |
| "middle": [], |
| "last": "\u8303\u9865\u9a30", |
| "suffix": "", |
| "affiliation": {}, |
| "email": "" |
| }, |
| { |
| "first": "", |
| "middle": [], |
| "last": "Fan", |
| "suffix": "", |
| "affiliation": {}, |
| "email": "" |
| } |
| ], |
| "year": "", |
| "venue": null, |
| "identifiers": {}, |
| "abstract": "The modulation spectra of speech features are often distorted due to environmental interferences. In order to reduce the distortion, in this paper we apply the minimum variance (MV) criterion to obtain the optimal frequency response of the temporal filter, and then two approaches, least-squares spectral fitting (LSSF) and magnitude spectrum interpolation (MSI) are used to obtain the filtered feature sequence. Accordingly, two new temporal processing approaches are proposed, which are named MV-LSSF and MV-MSI, respectively.", |
| "pdf_parse": { |
| "paper_id": "O10-1006", |
| "_pdf_hash": "", |
| "abstract": [ |
| { |
| "text": "The modulation spectra of speech features are often distorted due to environmental interferences. In order to reduce the distortion, in this paper we apply the minimum variance (MV) criterion to obtain the optimal frequency response of the temporal filter, and then two approaches, least-squares spectral fitting (LSSF) and magnitude spectrum interpolation (MSI) are used to obtain the filtered feature sequence. Accordingly, two new temporal processing approaches are proposed, which are named MV-LSSF and MV-MSI, respectively.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Abstract", |
| "sec_num": null |
| } |
| ], |
| "body_text": [ |
| { |
| "text": "In the Aurora-2 clean-condition training task, we show that the new MV-LSSF and MV-MSI give more than 50% relative error rate reduction over the baseline, and provide relative error rate reductions of 8.18% and 2.73% over the conventional LSSF and MSI, respectively. These results reveal that the proposed methods significantly enhance the robustness of speech features in noise-corrupted environments.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "Keywords: speech recognition, minimum variance, modulation spectra, robust speech features \u4e00\u3001\u7c21\u4ecb \u7e31\u4f7f\u8a9e\u97f3\u79d1\u6280\u65e5\u65b0\u6708\uf962\uff0c\u81ea\u52d5\u8a9e\u97f3\u8fa8\uf9fc(automatic speech recognition, ASR) [ ", |
| "cite_spans": [ |
| { |
| "start": 149, |
| "end": 150, |
| "text": "[", |
| "ref_id": null |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "\u95dc\u9375\u8a5e\uff1a\u81ea\u52d5\u8a9e\u97f3\u8fa8\uf9fc\u3001\u6700\u5c0f\u8b8a\uf962\uf969\u3001\u8abf\u8b8a\u983b\u8b5c\u3001\u5f37\u5065\u6027\u8a9e\u97f3\u7279\u5fb5", |
| "sec_num": null |
| }, |
| { |
| "text": "EQUATION", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [ |
| { |
| "start": 0, |
| "end": 8, |
| "text": "EQUATION", |
| "ref_id": "EQREF", |
| "raw_str": "( ) ( ) ( ) ( ) 1 2 1 2 L L j l l H hle w w - -- - = = \u00e5 \u5176\u4e2d\uff0cL \u70ba\uf984\u6ce2\u5668\u4fc2\uf969 ( ) h l \u7684\u9ede\uf969\uff0c\u70ba\u4e00\u5947\uf969\uff0c\u4e14 ( ) h l \u6eff\u8db3\u524d\u5f8c\u5c0d\u7a31\u7684\u6027\u8cea\uff0c \u5373 ( ) ( ) h l h l = -\u3002\u5c07\u5f0f(2)\u4ee3\u5165\u5f0f(1)\uff0c\u53ef\u5f97: ( ) ( ) ( ) 1 1 1 2 2 2 * 1 1 1 2 2 2 = ()() () L L L j l k j k N S L L L k l k h k h l P e d h k P e dw p p w w p p a l w w w - - - - - - - - - - =- =- =- - \u00e5 \u00e5 \u00e5 \u00f2 \u00f2 ( ) ( ) ( ) 1 1 1 2 2 2 * * 1 1 1 2 2 2 ( ) ( ) ( ) L L L j l k j l S S L L L l k l h l P e dw h k h l P e d p p w w p p w w w - - - - - - - - - =- =- =- - + \u00e5 \u00e5 \u00e5 \u00f2 \u00f2 ( ) S P dw p p w - + \u00f2 \u5f0f(3) \u85c9\u7531\u77e9\u9663\u8207\u5411\uf97e\u8868\u793a\u6cd5\uff0c\u5f0f(3)\u53ef\u6539\u5beb\u70ba: ( ) - 2 T T T N S S S P d p p a l w w = - + + \u00f2 h R h h r h R h \u5f0f(4) \u5176\u4e2d\uff0c N R \u70ba\u96dc\u8a0a\u5012\u983b\u8b5c\u7279\u5fb5\u7684\u81ea\u76f8\u95dc\u77e9\u9663(autocorrelation matrix)\uff0c S r \u70ba\u4e7e\u6de8\u8a9e\u97f3\u5012\u983b \u8b5c\u7279\u5fb5\u7684\u81ea\u76f8\u95dc\u4fc2\uf969\u5411\uf97e\uff0c S R \u70ba\u4e7e\u6de8\u8a9e\u97f3\u5012\u983b\u8b5c\u7279\u5fb5\u7684\u81ea\u76f8\u95dc\u77e9\u9663\uff0ch \u70ba\u8abf\u8b8a\uf984\u6ce2\u5668 \u4fc2\uf969\u5411\uf97e\uff0c\u5373 ( ) ( ) ( ) ( ) 1 2 1 2 h L h L \u00e9 \u00f9 = -- - \u00ea \u00fa \u00eb \u00fb h \uf04c \u3002 \u6211\u5011\u77e5\u9053\uff0c\u4e00\u542b\u6709\u52a0\u6210\u6027\u96dc\u8a0a\u7684\u8a9e\u97f3\u5012\u983b\u8b5c\u7279\u5fb5\uff0c\u5176\u96dc\u8a0a\u6210\u5206\u8207\u4e7e\u6de8\u8a9e\u97f3\u6210\u5206\u6210\u975e\u7dda\u6027 \u95dc\u4fc2\u7684\u7d44\u5408\uff0c\u4f46\u5728\u9019\uf9e8\uff0c\u70ba\uf9ba\ufa09\u4f4e\u8a08\u7b97\u7684\u8907\u96dc\ufa01\uff0c\u5047\u8a2d\u4e7e\u6de8\u8a9e\u97f3\u8207\u96dc\u8a0a\u4e8c\u6210\u5206\u5728\u76f8\u95dc\u4fc2 \uf969\u4e0a\u5448\u73fe\u7dda\u6027\u76f8\u52a0\uff1a N S N S + = + R R R , \u5f0f", |
| "eq_num": "(5" |
| } |
| ], |
| "section": "\u95dc\u9375\u8a5e\uff1a\u81ea\u52d5\u8a9e\u97f3\u8fa8\uf9fc\u3001\u6700\u5c0f\u8b8a\uf962\uf969\u3001\u8abf\u8b8a\u983b\u8b5c\u3001\u5f37\u5065\u6027\u8a9e\u97f3\u7279\u5fb5", |
| "sec_num": null |
| }, |
| { |
| "text": "EQUATION", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [ |
| { |
| "start": 0, |
| "end": 8, |
| "text": "EQUATION", |
| "ref_id": "EQREF", |
| "raw_str": "( ) - (1 ) 2 T T T N S S S S P d p p a l l w w + = + - - + \u00f2 h R h h R h h r . \u5f0f(6) 3. \u70ba\u6c42\u5f97\u6700\u4f73\uf984\u6ce2\u5668\u4fc2\uf969 h \u4f7f\u74b0\u5883\u5931\u771f\u9054\u5230\u6975\u5c0f\u503c\uff0c\u9019\uf9e8\u5c0d\u65bc\u4e0a\u5f0f(6)\u4f5c\u91dd\u5c0d h \u7684\u504f\u5fae \u5206\u4e26\uf9a8\u6046\u7b49\u5f0f\u70ba 0\uff0c\u53ef\u5f97\u6700\u4f73\u4e4b\u8abf\u8b8a\uf984\u6ce2\u5668\u4fc2\uf969\u70ba ( ) 1 (1 ) N S S S l l - + = + - h R R r . \u5f0f(7)", |
| "eq_num": "(2)" |
| } |
| ], |
| "section": "\u95dc\u9375\u8a5e\uff1a\u81ea\u52d5\u8a9e\u97f3\u8fa8\uf9fc\u3001\u6700\u5c0f\u8b8a\uf962\uf969\u3001\u8abf\u8b8a\u983b\u8b5c\u3001\u5f37\u5065\u6027\u8a9e\u97f3\u7279\u5fb5", |
| "sec_num": null |
| }, |
| { |
| "text": "EQUATION", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [ |
| { |
| "start": 0, |
| "end": 8, |
| "text": "EQUATION", |
| "ref_id": "EQREF", |
| "raw_str": "+ = , \u5f0f", |
| "eq_num": "(9)" |
| } |
| ], |
| "section": "\u95dc\u9375\u8a5e\uff1a\u81ea\u52d5\u8a9e\u97f3\u8fa8\uf9fc\u3001\u6700\u5c0f\u8b8a\uf962\uf969\u3001\u8abf\u8b8a\u983b\u8b5c\u3001\u5f37\u5065\u6027\u8a9e\u97f3\u7279\u5fb5", |
| "sec_num": null |
| }, |
| { |
| "text": "2. \u5047\u8a2d\u4e7e\u6de8\u8a9e\u97f3\u8207\u96dc\u8a0a\u4e8c\u6210\u5206\u5728\u529f\uf961\u983b\u8b5c\u5bc6\ufa01\u4e0a\u5448\u73fe\u7dda\u6027\u76f8\u52a0\uff0c\u5982\u4e0b\u6240\u793a: ", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "\u95dc\u9375\u8a5e\uff1a\u81ea\u52d5\u8a9e\u97f3\u8fa8\uf9fc\u3001\u6700\u5c0f\u8b8a\uf962\uf969\u3001\u8abf\u8b8a\u983b\u8b5c\u3001\u5f37\u5065\u6027\u8a9e\u97f3\u7279\u5fb5", |
| "sec_num": null |
| }, |
| { |
| "text": "EQUATION", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [ |
| { |
| "start": 0, |
| "end": 8, |
| "text": "EQUATION", |
| "ref_id": "EQREF", |
| "raw_str": "( ) ( ) ( ) N S N S P P P w w w + = + \u5f0f(10) \u5176\u4e2d\uff0c ( ) N S P w + \u70ba\u542b\u6709\u96dc\u8a0a\u8a9e\u97f3\u4e4b\u7279\u5fb5\u7684\u529f\uf961\u983b\u8b5c\u5bc6\ufa01\u3002\u4ee3\u5165\u5f0f(9)\uff0c\u53ef\u5f97\u6700\u4f73\uf984\u6ce2\u5668\u7684 \u983b\uf961\u97ff\u61c9\u5982\u4e0b\u6240\u793a: ( ) ( ) ( ) * ( ) 1 ( ) S N S S P H P P w w l w l w + = + - \u5f0f(11) \u7531\u4e0a\u5f0f\u53ef\u770b\u51fa ( ) * H w \u5176\u5be6\u662f\u4e00\u6b63\u5be6\uf969\uff0c\u6545\u53ef\u53bb\u6389\u8907\uf969\u5171\u8edb\u7b26\u865f\u800c\u5f97: ( ) ( ) ( ) ( ) 1 ( ) S N S S P H P P w w l w l w + = + - \u5f0f", |
| "eq_num": "(12" |
| } |
| ], |
| "section": "\u95dc\u9375\u8a5e\uff1a\u81ea\u52d5\u8a9e\u97f3\u8fa8\uf9fc\u3001\u6700\u5c0f\u8b8a\uf962\uf969\u3001\u8abf\u8b8a\u983b\u8b5c\u3001\u5f37\u5065\u6027\u8a9e\u97f3\u7279\u5fb5", |
| "sec_num": null |
| }, |
| { |
| "text": "\u5728 MV-LSSF \u6cd5\u4e2d\uff0c\u6211\u5011\u5c07\u6bcf\u4e00\u500b\u5f85\u6b63\u898f\u5316\u7684 N \u9ede\u7279\u5fb5\u5e8f\uf99c { } ;1 x n n N \u00e9 \u00f9 \u00a3 \u00a3 \u00ea \u00fa \u00eb \u00fb \u5148\u5b9a\u7fa9\u4e00 2P \u9ede\u7684\uf96b\u8003\u8abf\u8b8a\u983b\u8b5c\uff0c\u505a\u70ba\u6b64\u7279\u5fb5\u5e8f\uf99c\u7684\u8abf\u8b8a\u983b\u8b5c\u6b63\u898f\u5316\u7684\u76ee\u6a19\uff0c\u5982\u4e0b\u6240\u793a\uff1a ( ) ( ) ( ) ( ) exp , 0 2 1 k k X k Y Y j k P w w q w = \u00a3\u00a3 - \u5f0f(14) \u5176\u4e2d\u7684\u5f37\ufa01\u6210\u5206 ( ) k Y w \u5982\u4e0b\u5f0f\u8868\u793a\uff1a ( ) ( ) ( ) ( ) ( ) ( ) ( ) , 0 2 1 ( ) 1 S k k k k k X k S k P Y H X X k P P P w w w w w l w l w = = \u00a3 \u00a3 - + - \u5f0f(15) \u5176\u4e2d ( ) k H w \u5373\u662f\u63a1\u7528\u5f0f(12)\u4e4b\u6839\u64da\u6700\u5c0f\u8b8a\uf962\uf969\u6e96\u5247\u6240\u6c42\u53d6\u4e4b\u6700\u4f73\u983b\uf961\u97ff\u61c9\uff0c\u800c\u5f37\ufa01\u6210\u4efd ( ) k X w \u548c \u76f8 \u89d2 \u6210 \u4efd ( ) k q w \u70ba { } x n \u00e9 \u00f9 \u00ea \u00fa \u00eb \u00fb \u7d93 \u904e 2P \u9ede \u4e4b \uf9ea \u6563 \u5085 \uf9f7 \uf96e \u8f49 \u63db (discrete", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "\u95dc\u9375\u8a5e\uff1a\u81ea\u52d5\u8a9e\u97f3\u8fa8\uf9fc\u3001\u6700\u5c0f\u8b8a\uf962\uf969\u3001\u8abf\u8b8a\u983b\u8b5c\u3001\u5f37\u5065\u6027\u8a9e\u97f3\u7279\u5fb5", |
| "sec_num": null |
| }, |
| { |
| "text": "\u5728 MV-MSI \u6cd5\u7684\u904e\u7a0b\u4e2d\uff0c\u9996\u5148\u70ba\u6bcf\u4e00\u500b\u5f85\u6b63\u898f\u5316\u7684 N \u9ede\u7279\u5fb5\u5e8f\uf99c\uff0c { } ;1 x n n N \u00e9 \u00f9 \u00a3 \u00a3 \u00ea \u00fa \u00eb \u00fb \u5b9a \u7fa9\u4e00\u500b N \u9ede\u7684\uf96b\u8003\u8abf\u8b8a\u983b\u8b5c\uff0c\u4f5c\u70ba\u6b64\u5fb5\u5e8f\uf99c\u4e4b\u8abf\u8b8a\u983b\u8b5c\u6b63\u898f\u5316\u4e4b\u76ee\u6a19\uff0c\u5982\u4e0b\u6240\u793a\uff1a ( ) ( ) ( ) ' ' ' ' ? ( ) exp , 0 1 X k k k Y Y j k N w w q w = \u00a3\u00a3 - \u5f0f(17) \u5176\u4e2d\u76f8\u4f4d\u6210\u4efd ( ) ' X k q w \u70ba { } x n \u00e9 \u00f9 \u00ea \u00fa \u00eb \u00fb \u7d93\u904e N \u9ede\u4e4b DFT \u800c\u5f97\uff0c\u7531\u65bc MV-LSSF \u6cd5\u4e2d\u539f\u59cb 2P \u9ede\u7684 \uf96b\u8003\u983b\u8b5c(\u5982\u5f0f(14))\u6240\u6db5\u84cb\u7684\u983b\uf961\u7bc4\u570d\u8207\u9019\uf9e8\u5f0f(17)\u6240\u6c42\u7684 ( ) ' k Y w \u983b\uf961\u7bc4\u570d\u76f8\u540c\uff0c\u56e0\u6b64\u6211 \u5011\uf9dd\u7528\u7dda\u6027\u5167\u63d2\u6cd5(linear interpolation)\u7684\u65b9\u5f0f\uff0c\u85c9\u7531\u5f0f(15)\u7576\u4e2d 2P \u9ede\u4e4b\u4ee5\u6700\u5c0f\u8b8a\uf962\uf969\u6e96 1 1 2 k N Y N k N w \u00ec \u00fc \u00ef \u00ef \u00ef \u00ef -- \u00a3 \u00a3 - \u00ed \u00fd \u00ef \u00ef \u00ef \u00ef \u00ee \u00fe \u3002\u6700\u5f8c\u53ef\u5f97\u5230 ( ) { } 0 1 k Y k N w \u00a2 \u00a2 \u00a3 \u00a3 -\u3002 \u63a5\u4e0b\uf92d\uff0c\u6211\u5011\u76f4\u63a5\u5c0d\u5f0f(17)\u4e2d\u7684 ( ) { } k Y w \u00a2 \u4f5c N \u9ede\u7684\u53cd\u5085\uf9f7\uf96e\u8f49\u63db(inverse discrete Fourier trans, IDFT)\uff0c\u6c42\u5f97\u65b0\u7684\u7279\u5fb5\u5e8f\uf99c { } y n \u00e9 \u00f9 \u00ea \u00fa \u00eb \u00fb \uff0c\u5982\u4e0b\u6240\u793a: ( ) 1 0 1 2 exp , 0 1 N k k nk y n Y j n N N N p w - \u00a2 \u00a2= ae \u00f6 \u00a2 \u00f7 \u00e7 \u00e9 \u00f9 \u00f7 = \u00a3 \u00a3- \u00e7 \u00f7 \u00ea \u00fa \u00eb \u00fb \u00e7 \u00f7 \u00e7 \u00e8 \u00f8 \u00e5 \u5f0f(19) \u4e0a\u5f0f(19)\u7684 { } y n \u00e9 \u00f9 \u00ea \u00fa \u00eb \u00fb \u5373\u70ba\u85c9\u7531 MV-MSI", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "\u95dc\u9375\u8a5e\uff1a\u81ea\u52d5\u8a9e\u97f3\u8fa8\uf9fc\u3001\u6700\u5c0f\u8b8a\uf962\uf969\u3001\u8abf\u8b8a\u983b\u8b5c\u3001\u5f37\u5065\u6027\u8a9e\u97f3\u7279\u5fb5", |
| "sec_num": null |
| } |
| ], |
| "back_matter": [], |
| "bib_entries": { |
| "BIBREF1": { |
| "ref_id": "b1", |
| "title": "Phase autocorrelation (PAC) derived robust speech features", |
| "authors": [ |
| { |
| "first": "S", |
| "middle": [], |
| "last": "Ikbal", |
| "suffix": "" |
| }, |
| { |
| "first": "H", |
| "middle": [], |
| "last": "Hermansky", |
| "suffix": "" |
| }, |
| { |
| "first": "H", |
| "middle": [], |
| "last": "Bourlard", |
| "suffix": "" |
| } |
| ], |
| "year": 2003, |
| "venue": "2003 International Conference on Acoustics, Speech and Signal Processing", |
| "volume": "", |
| "issue": "", |
| "pages": "133--136", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "S. Ikbal, H. Hermansky and H. Bourlard, \"Phase autocorrelation (PAC) derived robust speech features,\" 2003 International Conference on Acoustics, Speech and Signal Processing (ICASSP 2003), pp.133-136, 2003.", |
| "links": null |
| }, |
| "BIBREF2": { |
| "ref_id": "b2", |
| "title": "New approaches for domain transformation and parameter combination for improved accuracy in parallel model combination (PMC) techniques", |
| "authors": [ |
| { |
| "first": "J", |
| "middle": [], |
| "last": "Hung", |
| "suffix": "" |
| }, |
| { |
| "first": "J", |
| "middle": [], |
| "last": "Shen", |
| "suffix": "" |
| }, |
| { |
| "first": "L", |
| "middle": [], |
| "last": "Lee", |
| "suffix": "" |
| } |
| ], |
| "year": 2001, |
| "venue": "IEEE Trans. on Speech and Audio Processing", |
| "volume": "", |
| "issue": "", |
| "pages": "842--855", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "J. Hung, J. Shen and L. Lee, \"New approaches for domain transformation and parameter combination for improved accuracy in parallel model combination (PMC) techniques,\" IEEE Trans. on Speech and Audio Processing, pp.842-855, 2001.", |
| "links": null |
| }, |
| "BIBREF3": { |
| "ref_id": "b3", |
| "title": "Cepstral analysis technique for automatic speaker verification", |
| "authors": [ |
| { |
| "first": "S", |
| "middle": [], |
| "last": "Furui", |
| "suffix": "" |
| } |
| ], |
| "year": 1981, |
| "venue": "IEEE Trans. on Acoustics, Speech and Signal Processing", |
| "volume": "", |
| "issue": "", |
| "pages": "254--272", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "S. Furui, \"Cepstral analysis technique for automatic speaker verification,\" IEEE Trans. on Acoustics, Speech and Signal Processing, pp.254-272, 1981.", |
| "links": null |
| }, |
| "BIBREF4": { |
| "ref_id": "b4", |
| "title": "Cepstral domain segmental feature vector normalization for noise robust speech recognition", |
| "authors": [ |
| { |
| "first": "O", |
| "middle": [], |
| "last": "Viikki", |
| "suffix": "" |
| }, |
| { |
| "first": "K", |
| "middle": [], |
| "last": "Laurila", |
| "suffix": "" |
| } |
| ], |
| "year": 1998, |
| "venue": "Proceedings of the 38th Southeastern Symposium on System Theory Speech Communication", |
| "volume": "25", |
| "issue": "", |
| "pages": "133--147", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "O. Viikki and K. Laurila, \"Cepstral domain segmental feature vector normalization for noise robust speech recognition,\" Proceedings of the 38th Southeastern Symposium on System Theory Speech Communication, Vol. 25, pp.133-147, 1998.", |
| "links": null |
| }, |
| "BIBREF5": { |
| "ref_id": "b5", |
| "title": "RASTA processing of speech features", |
| "authors": [ |
| { |
| "first": "H", |
| "middle": [], |
| "last": "Hermansky", |
| "suffix": "" |
| }, |
| { |
| "first": "N", |
| "middle": [], |
| "last": "Morgan", |
| "suffix": "" |
| } |
| ], |
| "year": 1994, |
| "venue": "IEEE Trans. on Speech and Audio Processing", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "H. Hermansky and N. Morgan, \"RASTA processing of speech features,\" IEEE Trans. on Speech and Audio Processing, 1994.", |
| "links": null |
| }, |
| "BIBREF6": { |
| "ref_id": "b6", |
| "title": "MVA processing of speech features", |
| "authors": [ |
| { |
| "first": "C", |
| "middle": [], |
| "last": "Chen", |
| "suffix": "" |
| }, |
| { |
| "first": "J", |
| "middle": [], |
| "last": "Bilmes", |
| "suffix": "" |
| } |
| ], |
| "year": 2006, |
| "venue": "Speech, and Language Processing", |
| "volume": "", |
| "issue": "", |
| "pages": "257--270", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "C. Chen and J. Bilmes, \"MVA processing of speech features,\" IEEE Trans. on Audio, Speech, and Language Processing, pp.257-270, 2006.", |
| "links": null |
| }, |
| "BIBREF7": { |
| "ref_id": "b7", |
| "title": "Quantile based histogram equalization for noise robust large vocabulary speech recognition", |
| "authors": [ |
| { |
| "first": "F", |
| "middle": [], |
| "last": "Hilger", |
| "suffix": "" |
| }, |
| { |
| "first": "H", |
| "middle": [], |
| "last": "Ney", |
| "suffix": "" |
| } |
| ], |
| "year": 2006, |
| "venue": "Speech and Language Processing", |
| "volume": "", |
| "issue": "", |
| "pages": "845--854", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "F. Hilger and H. Ney, \"Quantile based histogram equalization for noise robust large vocabulary speech recognition,\" IEEE Trans. on Audio, Speech and Language Processing, pp.845-854, 2006.", |
| "links": null |
| }, |
| "BIBREF8": { |
| "ref_id": "b8", |
| "title": "Normalizing the speech modulation spectrum for robust speech recognition", |
| "authors": [ |
| { |
| "first": "X", |
| "middle": [], |
| "last": "Xiao", |
| "suffix": "" |
| }, |
| { |
| "first": "E", |
| "middle": [ |
| "S" |
| ], |
| "last": "Chug", |
| "suffix": "" |
| }, |
| { |
| "first": "H", |
| "middle": [], |
| "last": "Li", |
| "suffix": "" |
| } |
| ], |
| "year": 2007, |
| "venue": "IEEE International Conference on Acoustics, Speech and Signal Processing", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "X. Xiao, E. S. Chug and H. Li, \"Normalizing the speech modulation spectrum for robust speech recognition,\" IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP 2007), 2007.", |
| "links": null |
| }, |
| "BIBREF9": { |
| "ref_id": "b9", |
| "title": "Improved modulation spectrum normalization techniques for robust speech recognition", |
| "authors": [ |
| { |
| "first": "C", |
| "middle": [ |
| "C" |
| ], |
| "last": "Pan", |
| "suffix": "" |
| }, |
| { |
| "first": "J", |
| "middle": [], |
| "last": "Wang", |
| "suffix": "" |
| }, |
| { |
| "first": "", |
| "middle": [], |
| "last": "Hung", |
| "suffix": "" |
| } |
| ], |
| "year": 2008, |
| "venue": "2008 International Conference on Acoustics, Speech and Signal Processing", |
| "volume": "", |
| "issue": "", |
| "pages": "4089--4092", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "C. Pan C. Wang J. Hung, \"Improved modulation spectrum normalization techniques for robust speech recognition,\" 2008 International Conference on Acoustics, Speech and Signal Processing (ICASSP 2008), pp.4089-4092, 2008.", |
| "links": null |
| }, |
| "BIBREF10": { |
| "ref_id": "b10", |
| "title": "Minimum variance modulation filter for robust speech recognition", |
| "authors": [ |
| { |
| "first": "Y", |
| "middle": [ |
| "B" |
| ], |
| "last": "Chiu", |
| "suffix": "" |
| }, |
| { |
| "first": "R", |
| "middle": [ |
| "M" |
| ], |
| "last": "Stern", |
| "suffix": "" |
| } |
| ], |
| "year": 2009, |
| "venue": "2009 International Conference on Acoustics, Speech and Signal Processing", |
| "volume": "", |
| "issue": "", |
| "pages": "3962--3965", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Y. B. Chiu and R. M. Stern, \"Minimum variance modulation filter for robust speech recognition,\" 2009 International Conference on Acoustics, Speech and Signal Processing (ICASSP 2009), pp.3962-3965, 2009.", |
| "links": null |
| }, |
| "BIBREF11": { |
| "ref_id": "b11", |
| "title": "The AURORA experimental framework for the performance evaluations of speech recognition systems under noisy conditions", |
| "authors": [ |
| { |
| "first": "H", |
| "middle": [ |
| "G" |
| ], |
| "last": "Hirsch", |
| "suffix": "" |
| }, |
| { |
| "first": "D", |
| "middle": [], |
| "last": "Pearce", |
| "suffix": "" |
| } |
| ], |
| "year": 2000, |
| "venue": "Proceedings of ISCA IIWR ASR2000", |
| "volume": "", |
| "issue": "", |
| "pages": "181--188", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "H. G. Hirsch and D. Pearce, \"The AURORA experimental framework for the performance evaluations of speech recognition systems under noisy conditions,\" Proceedings of ISCA IIWR ASR2000, pp.181-188, 2000.", |
| "links": null |
| }, |
| "BIBREF12": { |
| "ref_id": "b12", |
| "title": "Transmission Performance Characteristics of Pulse Code Modulation Channels", |
| "authors": [], |
| "year": 1996, |
| "venue": "", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "ITU recommendation G.712, Transmission Performance Characteristics of Pulse Code Modulation Channels, Nov. 1996.", |
| "links": null |
| } |
| }, |
| "ref_entries": { |
| "TABREF0": { |
| "content": "<table><tr><td>1]\u4f9d\u820a\u662f\u773e\u591a \u5c08\u5bb6\u3001\u5b78\u8005\u7814\u7a76\u958b\u767c\u7684\u6a19\u7684\u3002\u4e3b\u8981\u539f\u56e0\u5728\u65bc\u5be6\u969b\u751f\u6d3b\u74b0\u5883\u4e2d\u5b58\u5728\u8457\u591a\u65b9\u9762\u7684\u8b8a\uf962\u6027 (variation)\u5f71\u97ff\u8fa8\uf9fc\u6548\u679c\uff0c\u9019\u7576\u4e2d\u5f71\u97ff\u8a9e\u97f3\u8fa8\uf9fc\u7684\u8b8a\uf962\u6027\u5305\u542b\uf9ba\u8a13\uf996\u74b0\u5883\u8207\u61c9\u7528\u74b0\u5883\u4e4b \u9593\u7684\u74b0\u5883\uf967\u5339\u914d(environmental mismatch)\u3001\u8a13\uf996\u8a9e\u8005\u8207\u61c9\u7528\u8a9e\u8005\u4e4b\u9593\u7684\u8a9e\u8005\u5dee\uf962\u6027 (speaker variation)\u53ca\uf967\u540c\u8a9e\u8005\u6216\u540c\u4e00\u8a9e\u8005\u5728\u767c\u97f3\u4e0a\u7684\u8b8a\uf962\u6027(pronunciation variation)\u7b49\u8a31 \u591a\u56e0\u7d20\uff0c\u9019\u4e9b\u56e0\u7d20\u90fd\u6703\u660e\u986f\u5f71\u97ff\u8a9e\u97f3\u8fa8\uf9fc\u7cfb\u7d71\u7684\u6548\u80fd\u3002\u56e0\u6b64\u5728\u8fd1\u5e7e\u5341\uf98e\uf92d\uff0c\u6709\u8a31\u8a31\u591a\u591a \u7684\u5b78\u8005\u6301\u7e8c\uf967\u65b7\u671d\u8457\u52aa\uf98a\u6539\u5584\u4ee5\u4e0a\u5e7e\u7a2e\u7684\u8a9e\u97f3\u5dee\uf962\u6027\uff0c\u9032\u800c\u4f7f\u8a9e\u97f3\u8fa8\uf9fc\u7cfb\u7d71\u80fd\uf901\u6709\u6548\u5730 \u904b\u7528\u65bc\u771f\u5be6\u7684\u751f\u6d3b\u74b0\u5883\u4e2d\u3002 \u91dd\u5c0d\u74b0\u5883\uf967\u5339\u914d\u7684\uf9fa\u6cc1\u767c\u5c55\u51fa\u8a31\u591a\u5f37\u5065\u6027\u65b9\u6cd5\uff0c\u7d9c\u89c0\u800c\u8a00\uff0c\u5927\u81f4\u5305\u542b\u7279\u5fb5\u88dc\u511f[2]\u8207\u6a21 \u578b\u88dc\u511f[3]\uf978\u5927\uf9d0\u578b\uff0c\u800c\u7279\u5fb5\u88dc\u511f\u65b9\u6cd5\u7576\u4e2d\uff0c\u6709\u4e00\uf9d0\u578b\u662f\u91dd\u5c0d\u8a9e\u97f3\u8fa8\uf9fc\u6240\u7528\u7684\u7279\u5fb5\uf96b\uf969 \u4e4b\u7d71\u8a08\uf97e\u4f5c\u6b63\u898f\u5316\u8655\uf9e4\uff0c\u9019\u4e9b\u8655\uf9e4\u65b9\u5f0f\u901a\u5e38\u662f\u4f5c\u5728\u7279\u5fb5\u4e4b\u6642\u9593\u5e8f\uf99c\u57df(temporal domain) \u4e0a\uff0c\u76ee\u7684\u662f\u5c07\u5f37\u8abf\u7279\u5fb5\u4e2d\u8a9e\u97f3\u7684\u6210\u5206\uff0c\u5c07\u96dc\u8a0a\u7684\u6210\u5206\u58d3\u6291\u4e0b\uf92d\uff0c\u6216\u662f\u4f7f\uf967\u540c\u74b0\u5883\u4e0b\u7684\u8a9e \u97f3\u7279\u5fb5\u4e4b\u7d71\u8a08\uf97e\u90fd\u80fd\u8da8\u65bc\u4e00\u81f4\uff0c\u85c9\u6b64\u63d0\u9ad8\u8a9e\u97f3\u8fa8\uf9fc\uf961\uff0c\u9019\u4e9b\u65b9\u6cd5\uf9b5\u5982\u5012\u983b\u8b5c\u5e73\u5747\u503c\u6b63\u898f \u5316\u6cd5(cepstral mean normalization, CMN)[4]\u3001\u5012\u983b\u8b5c\u5e73\u5747\u503c\u8207\u8b8a\uf962\uf969\u6b63\u898f\u5316\u6cd5(cepstral mean and variance normalization, CMVN)[5]\u3001\u76f8\u5c0d\u983b\u8b5c\u6cd5(RelAtive SpecTra, RASTA)[6]\u3001 \u5012\u983b\u8b5c\u5e73\u5747\u503c\u8207\u8b8a\uf962\uf969\u6b63\u898f\u5316\u7d50\u5408\u81ea\u56de\u6b78\u52d5\u614b\u5e73\u5747\uf984\u6ce2\u5668\u6cd5(cepstral mean and variance normalization plus auto-regressive-moving average filtering, MVA)[7]\u3001\u8207\u7d71\u8a08\u5716\u7b49\u5316\u6cd5 (histogram equalization, HEQ)[8]\u3001\u6642\u9593\u5e8f\uf99c\u7d50\u69cb\u6b63\u898f\u5316\u6cd5(temporal structure normalization, TSN)[9]\u3001\u7b49\uf992\u6ce2\u6642\u9593\u5e8f\uf99c\uf984\u6ce2\u5668\u6cd5(equi-ripple temporal filter, ERTF)[10]\u3001\u6700\u5c0f\u5e73\u65b9\u983b\u8b5c \u64ec\u5408\u6cd5(least squares spectrum fitting, LSSF) [10]\u3001\u5f37\ufa01\u983b\u8b5c\u5167\u63d2\u6cd5(magnitude spectrum interpolation, MSI) [10]\u7b49\u3002\u800c\u7279\u5225\u4e00\u63d0\u7684\u662f\uff0c2009 \uf98e\u6642\u6709\u5b78\u8005\u63d0\u51fa\uf9ba\u6700\u5c0f\u8b8a\uf962\uf969\u8abf\u8b8a \uf984\u6ce2\u5668\u8a2d\u8a08\u6cd5 (minimum variance modulation filter, MVMF)[11]\uff0c\u5176\u4e3b\u8981\u662f\u6839\u64da\u6700\u5c0f\u5316\u96dc \u8a0a\u8b8a\uf962\uf969\u7684\u6700\u4f73\u5316\u76ee\u6a19\uff0c\u9032\u800c\u63a8\u5f97\u7279\u5fb5\u4e4b\u6642\u9593\u5e8f\uf99c\u57df\u4e0a\u7684\uf984\u6ce2\u5668\u8108\u885d\u97ff\u61c9(impulse response)\uff0c\u85c9\u7531\u5c0d\u8a9e\u97f3\u7279\u5fb5\uf984\u6ce2\u8655\uf9e4\uff0c\u800c\u6539\u5584\u8a9e\u97f3\u7279\u5fb5\u7684\u96dc\u8a0a\u5f37\u5065\u6027\u3002 \u4ee5\u4e0a\u5404\u7a2e\u6280\u8853\u4e3b\u8981\u662f\u76f4\u63a5\u6216\u9593\u63a5\u57f7\ufa08\u5728\u8a9e\u97f3\u7279\u5fb5\u7684\u6642\u9593\u5e8f\uf99c\u57df\u4e0a\uff0c\u4f46\u5728\u5176\u6548\u80fd\u7684\u5206\u6790 \u4e0a\uff0c\u6211\u5011\u901a\u5e38\u6703\u53bb\u63a2\u8a0e\u96dc\u8a0a\u53ca\u901a\u9053\u6548\u61c9\u5c0d\u65bc\u539f\u59cb\u7279\u5fb5\u4e4b\u8abf\u8b8a\u983b\u8b5c\u7684\u5931\u771f\uff0c\u53ca\u9019\u4e9b\u65b9\u6cd5\u5c0d \u65bc\u6b64\u5931\u771f\u7684\u6539\u5584\u7a0b\ufa01\uff0c \u56e0\u6b64\u5728\u672c\u7bc7\uf941\u6587\u4e2d\uff0c\u6211\u5011\uf96b\u8003\uf9ba MVMF \u6cd5[11]\u7684\u69cb\u60f3\uff0c\u4f7f\u7528\u8b8a \uf962\uf969\u6700\u5c0f\u5316\u4e4b\u6700\u4f73\u6e96\u5247\uf92d\u8655\uf9e4\u8a9e\u97f3\u7279\u5fb5\uff0c\u4f46\u6211\u5011\u6240\u767c\u5c55\u7684\u65b9\u6cd5\u8207\u539f\u59cb MVMF \u6cd5\uf967\u540c\u9ede \u5728\u65bc\uff0c\u5b83\u5011\u662f\u6c42\u5f97\u6700\u4f73\u7684\uf984\u6ce2\u5668\u4e4b\u983b\uf961\u97ff\u61c9(frequency response)\uff0c\u5373\u8abf\u8b8a\u983b\u57df\u4e0a\u7684\u6700\u4f73 MV-MSI \u6cd5\uff1b\u7b2c\u4e09\u7ae0\u5c07\u5448\u73fe\u672c\uf941\u6587\u6240\u63d0\u51fa\u7684\u65b0\u65b9\u6cd5\u4e4b\u8fa8\uf9fc\u5be6\u9a57\u7d50\u679c\u8207\u8a0e\uf941\uff0c\u7b2c\u56db\u7ae0\u70ba \u7d50\uf941\u8207\u672a\uf92d\u5c55\u671b\u3002 \u4e8c\u3001\u57fa\u65bc\u6700\u5c0f\u8b8a\uf962\uf969\u4e4b\u8abf\u8b8a\u983b\u8b5c\u6b63\u898f\u5316\u6cd5 \u5728\u672c\u7ae0\u4e2d\uff0c\u6211\u5011\u9996\u5148\u7c21\uf976\u4ecb\u7d39\u524d\u5b78\u8005\u6240\u63d0\u51fa\u4e4b\u6700\u5c0f\u8b8a\uf962\uf969\u8abf\u8b8a\uf984\u6ce2\u5668\u6cd5[11]\uff0c\u63a5\u8457\uff0c\u6211 \u5011\u4ecb\u7d39\u672c\uf941\u6587\u6240\u63d0\u51fa\u4e4b\uf978\u500b\u65b0\u65b9\u6cd5\uff0c\u5373\u662f\u5c07\u6700\u5c0f\u8b8a\uf962\uf969\u8abf\u8b8a\uf984\u6ce2\u5668\u8a2d\u8a08\u7684\u76ee\u6a19\u51fd\uf969\uff0c\u61c9 \u7528\u81f3\u6c42\u53d6\uf984\u6ce2\u5668\u4e4b\u983b\uf961\u97ff\u61c9\u4e0a\uff0c\u9032\u800c\u5ef6\u4f38\u51fa\uf978\u7a2e\u7279\u5fb5\u6642\u9593\u5e8f\uf99c\u8655\uf9e4\u6f14\u7b97\u6cd5\uff0c\u5206\u5225\u70ba\u57fa\u65bc \u6700\u5c0f\u8b8a\uf962\uf969\u4e4b\u6700\u5c0f\u5e73\u65b9\uf984\u6ce2\u5668\u6cd5(MV-LSSF)\u8207\u53ca\u57fa\u65bc\u6700\u5c0f\u8b8a\uf962\uf969\u4e4b\u5f37\ufa01\u983b\u8b5c\u5167\u63d2\u6cd5 (MV-MSI)\uff0c\u9019\uf978\u7a2e\u65b0\u65b9\u6cd5\u7684\u8a73\u7d30\u6b65\u9a5f\u5c07\u65bc\u672c\u7ae0\u8a73\u8ff0\u3002 (1)\u6700\u5c0f\u8b8a\uf962\uf969\u8abf\u8b8a\uf984\u6ce2\u5668\u6cd5(minimum variance modulation filter, MVMF) \u4e00\u8a2d\u8a08\u5f97\u7576\u7684\u7279\u5fb5\u6642\u9593\u5e8f\uf99c\uf984\u6ce2\u5668\uff0c\u53ef\u4ee5\u51f8\u986f\u7279\u5fb5\u4e2d\u7684\u8a9e\u97f3\u6210\u5206\u4e26\u6291\u5236\u96dc\u8a0a\u6210\u5206\uff0c\u9032\u800c \u63d0\u5347\u8a9e\u97f3\u7279\u5fb5\u7684\u5f37\u5065\u6027\u3002\u800c\u6700\u5c0f\u8b8a\uf962\uf969\uf984\u6ce2\u5668(MVMF)\u8a2d\u8a08\u6cd5[11]\uff0c\u4e3b\u8981\u662f\u6839\u64da\u4e09\u500b\u65b9 \u5411\uf92d\u8a2d\u8a08\u7279\u5fb5\u7684\u6642\u9593\u5e8f\uf99c\uf984\u6ce2\u5668\uff1a 1. \uf984\u6ce2\u5668\u672c\u8eab\u53ef\u4ee5\u96a8\u8457\uf967\u540c\u8a9e\uf906(\u53ef\u80fd\u5c0d\u61c9\uf967\u540c\u7684\u96dc\u8a0a\u5e72\u64fe\u74b0\u5883)\u800c\u4f5c\u52d5\u614b\u8abf\u6574\u3002 2. \u5b9a\u7fa9\u4e00\u500b\u300e\u74b0\u5883\u5931\u771f\u300f\u7684\u76ee\u6a19\u51fd\uf969\uff0c\u7d93\u7531\u8abf\u8b8a\uf984\u6ce2\u5668\u7684\u8a2d\u8a08\uff0c\u4f7f\u8655\uf9e4\u5f8c\u7684\u74b0\u5883\u5931\u771f \u76ee\u6a19\u51fd\uf969\u503c\u80fd\u8da8\u65bc\u6700\u5c0f\u3002 3. \uf984\u6ce2\u5668\u7684\u8a2d\u8a08\uff0c\u9664\uf9ba\u8003\uf97e\u5230\ufa09\u4f4e\u96dc\u8a0a\u6210\u5206\u5916\uff0c\u4e5f\u540c\u6642\u8003\u616e\u5230\u539f\u59cb\u8a9e\u97f3\u6210\u5206\u5118\uf97e\uf967\u53d7 \u5230\u5f71\u97ff\u8207\uf901\u52d5\u3002 \u9644\u5e36\u4e00\u63d0\u7684\u662f\uff0c\u5728\u6587\u737b[11]\u4e2d\u7684\u6700\u5c0f\u8b8a\uf962\uf969\u8abf\u8b8a\uf984\u6ce2\u5668\u70ba\u7b2c\u4e00\uf9d0\u7dda\u6027\u76f8\u4f4d\uf984\u6ce2\u5668(type I linear-phase filter)\uff0c\u5373\uf984\u6ce2\u5668\u9577\ufa01\u70ba\u5947\uf969\u4e14\u524d\u5f8c\u5c0d\u7a31\uff0c\u5982\u6211\u5011\u6240\u77e5\uff0c\u7dda\u6027\u76f8\u4f4d\uf984\u6ce2\u5668\u53ea \u6539\u8b8a\u8f38\u5165\u8a0a\u865f\u7684\u983b\u8b5c\u5f37\ufa01\u53ca\u9020\u6210\u56fa\u5b9a\u7684\u6642\u9593\u5ef6\u9072\uff0c\u4e26\uf967\u6703\u9020\u6210\u8a0a\u865f\u6642\u9593\u5ef6\u9072\u4e0a\u7684\u5931\u771f\uff0c \u800c\u7b2c\u4e00\uf9d0\u7dda\u6027\u76f8\u4f4d\uf984\u6ce2\u5668\u984d\u5916\u512a\u9ede\u5247\u662f\u53ef\u4ee5\uf967\u53d7\u9650\u5730\u8fd1\u4f3c\u5404\u7a2e\u578b\u614b\u7684\uf984\u6ce2\u5668(\u5982\u4f4e\u901a\u3001 \u9ad8\u901a\u3001\u5e36\u901a\u8207\u5e36\u62d2\u7b49\uf984\u6ce2\u5668\u7b49) \u3002 \u4ee5\u4e0b\u70ba MVMF \u6cd5\u4e2d\u8a2d\u8a08\uf984\u6ce2\u5668\u4fc2\uf969\u7684\u6b65\u9a5f\uff1a 1. \u5c0d\u65bc\u4efb\u4e00\u8a9e\uf906\u7684\u67d0\u4e00\u7dad\u7279\u5fb5\u6642\u9593\u5e8f\uf99c\uff0c\u5b9a\u7fa9\u5176\u300e\u74b0\u5883\u5931\u771f\u300f(environmental mismatch) \u5982\u4e0b\u5f0f\uff1a ( ) ( ) ( ) ( ) 2 2 --1 -N S H P d H P d p p p p a l w w w w w w = + \u00f2 \u00f2 \u5f0f(1) \u5316\uff0c\u518d\u7d93\u7531\u524d\u8ff0\u4e4b LSSF \u8207 \u672c\uf941\u6587\u5176\u4ed6\u7ae0\u7bc0\u6982\u8981\u5982\u4e0b\uff1a\u5728\u7b2c\u4e8c\u7ae0\u4e2d\u4ecb\u7d39\u672c\uf941\u6587\u6240\u63d0\u51fa\u4e4b\u65b0\u65b9\u6cd5\uff0c\u5373 MV-LSSF \u8207 \u5176\u4e2d\uff0c ( ) H w \u70ba\uf984\u6ce2\u5668\u4e4b\u983b\uf961\u97ff\u61c9\uff0c ( ) N P w \u70ba\u96dc\u8a0a\u7684\u529f\uf961\u983b\u8b5c\u5bc6\ufa01\uff0c ( )</td></tr></table>", |
| "html": null, |
| "type_str": "table", |
| "num": null, |
| "text": "MSI \u6cd5\uff0c\u6c42\u5f97\uf984\u6ce2\u5668\u8655\uf9e4\u5f8c\u7684\u8a9e\u97f3\u7279\u5fb5\uff0c\u800c\u4e26\u975e\u5982 MVMF \u6cd5\u76f4\u63a5\u5728\u7279\u5fb5\u4e4b\u6642\u9593\u5e8f\uf99c\u57df\u7684\u6700\u4f73\u5316\u6c42\u5f97\uf984\u6ce2\u5668\u7684\u8108\u885d\u97ff\u61c9\u3002\u5728\u5be6\u9a57\u7d50\u679c\u767c\u73fe\uff0c\u6211\u5011\u6240 \u65b0\u63d0\u51fa\u7684 MV-LSSF \u6cd5\u8207 MV-MSI \u6cd5\uff0c\u6240\u5c0d\u61c9\u4e4b\u8fa8\uf9fc\u6548\u80fd\u512a\u65bc\u539f\u59cb LSSF \u6cd5\u8207 MSI \u6cd5\uff0c\u4e14 MV-LSSF \u6cd5\u512a\u65bc MVMF \u6cd5\uff0c\u800c MV-MSI \u6cd5\u6548\u679c\u5247\u8207 MVMF \u6cd5\u5341\u5206\u76f8\u8fd1\u3002" |
| }, |
| "TABREF5": { |
| "content": "<table><tr><td>\u6240\u5f97\u4e4b\u65b0\u8a9e\u97f3\u7279\u5fb5\u5e8f\uf99c\u3002 \u5728\u9019\uf9e8\uff0c\u6211\u5011\u5c0d\u63d0\u51fa\u7684 MV-LSSF \u8207 MV-MSI \u6cd5\u8207 MVMF \u6cd5\u4f5c\u521d\u6b65\u6548\u80fd\u7684\u6bd4\u8f03\uff0c\u6211\u5011 \u6839\u64da\u9019\u4e9b\u65b9\u6cd5\u5728\u8a9e\u97f3\u7279\u5fb5\u5e8f\uf99c\u4e4b\u8abf\u8b8a\u983b\u8b5c\u7684\u5931\u771f\u6539\u5584\u7a0b\ufa01\uff0c\uf92d\u8a55\u4f30\u9019\u4e9b\u65b9\u6cd5\u7684\u6548\u80fd\u3002\u9019 \uf9e8\u4f7f\u7528 AURORA-2 \u8cc7\uf9be\u5eab[12]\u4e2d\u7684 FAK_3Z82A \u4e7e\u6de8\u8a9e\u97f3\u6a94\uff0c\u7d93\u7531\u52a0\u5165\uf967\u540c\u8a0a\u96dc\u6bd4 (signal-to-noise ratio, SNR)\u7684\u5730\u4e0b\u9435(subway)\u96dc\u8a0a\u6240\u7522\u751f\u4e4b\u96dc\u8a0a\u8a9e\u97f3\u6a94\uff0c\u8f49\u6210\u7279\u5fb5\u51fd\uf969 \u5f8c\uff0c\u518d\u5206\u5225\u7d93\u904e\u4e0a\u8ff0\u4e09\u7a2e\u65b9\u6cd5\u4f5c\u8655\uf9e4\u3002\u5716\u4e00(a)(b)(c)(d)\u5206\u5225\u4ee3\u8868\u539f\u59cb\u672a\u7d93\u8655\uf9e4\u4e4b\u7b2c\u5341\u4e8c \u7dad MFCC \u7279\u5fb5\u5e8f\uf99c(c 12 )\u3001MVMF \u6cd5\u3001MV-LSSF \u6cd5\u8207 MV-MSI \u6cd5\u8655\uf9e4\u5f8c\u4e4b c 12 \u5e8f\uf99c\u7684 \u529f\uf961\u983b\u8b5c\u5bc6\ufa01(power spectral density, PSD)\u66f2\u7dda\u5716\u3002\u6839\u64da\u5716\u4e00\uff0c\u6211\u5011\u53ef\u4ee5\u767c\u73fe\uff1a 1. \u89c0\u5bdf\u5716\u4e00(a)\u5f97\u77e5\uff0c\u5728\uf967\u540c SNR \u503c\u4e0b(clean, 10dB \u8207 0dB)\uff0c\u672a\u8655\uf9e4\u904e\u7684 c 12 \u5e8f\uf99c\uff0c\u5176\u529f \uf961\u983b\u8b5c\u5bc6\ufa01(PSD)\u66f2\u7dda\u56e0\u53d7\u5230\u52a0\u6210\u6027\u96dc\u8a0a(additive noise)\u7684\u5f71\u97ff\uff0c\u5b58\u5728\u660e\u986f\u7684\u5931\u771f\u60c5\u5f62\u3002 \u800c\u7d93\u7531\u5716\u4e00(b)\u53ef\u770b\u51fa\uff0cMVMF \u6cd5\u8655\uf9e4\u5f8c\u4e4b c 12 \u5e8f\uf99c\uff0c\u5728\u8f03\u4f4e\u7684\u8abf\u8b8a\u983b\uf961\u7bc4\u570d[0,10Hz]\uff0c \u5176 PSD \u5931\u771f\u7684\u60c5\u6cc1\u5df2\u6709\u5f88\u660e\u986f\u7684\ufa09\u4f4e\uff0c\u4f46\u76f8\u5c0d\u65bc\u8f03\u9ad8\u7684\u8abf\u8b8a\u983b\uf961\u7bc4\u570d\uff0cPSD \u5931\u771f\u7684\u60c5 \u5f62\u4e26\u6c92\u6709\u592a\u5927\u7684\u6539\u5584\u3002 2. \u5716\u4e00(c)\u70ba MV-LSSF \u6cd5\u8655\uf9e4\u5f8c\u6240\u5f97\u5230\u7684 c 12 \u5e8f\uf99c\u4e4b PSD \u5716\uff0c\u5f88\u660e\u986f\u53ef\u770b\u51fa\u5728\u5168\u90e8\u7684\u8abf \u8b8a\u983b\uf961\u7bc4\u570d\uff0c\u5176 PSD \u5931\u771f\u60c5\u5f62\u7686\u6709\u6548\u5730\ufa09\u4f4e\uff0c\u5c24\u5176\u662f\u8f03\u9ad8\u983b\u7684\u8abf\u8b8a\u983b\uf961\u7bc4\u570d[35, 50Hz]\uff0c\u5728\uf967\u540c\u7684 SNR \u503c\u4e0b\u4e4b PSD \u66f2\u7dda\u8da8\u8fd1\u65bc\u4e7e\u6de8\u8a9e\u97f3\u7279\u5fb5(clean)\u4e4b PSD \u66f2\u7dda\uff1b\u800c \u5716\u4e00(d)\u70ba MV-MSI \u6cd5\u8655\uf9e4\u5f8c\u6240\u5f97\u5230\u7684\u7279\u5fb5\u5e8f\uf99c\u4e4b PSD \u5716\uff0c\u53ef\u770b\u51fa\uf967\u7ba1\u8f03\u4f4e\u7684\u8abf\u8b8a\u983b \uf961[0,10Hz]\u4e4b\u9593\u6216\u8f03\u9ad8\u7684\u8abf\u8b8a\u983b\uf961\u7bc4\u570d[10Hz, 50Hz]\uff0c\u5176 PSD \u5931\u771f\u7684\u60c5\u6cc1\u4e5f\u6709\u5f88\u660e\u986f\u7684 \ufa09\u4f4e\u3002 \u5716\u4e00\uff1a (a)\u539f\u59cb c 12 \u7279\u5fb5\u5e8f\uf99c\u3001(b)MVMF \u6cd5\u3001(c)MV-LSSF \u6cd5\u8207(d)MV-MSI \u6cd5\u4f5c\u7528\u65bc\uf967 \u540c\u8a0a\u96dc\u6bd4\u4e0b\u4e4b c 12 \u7279\u5fb5\u5e8f\uf99c\u4e4b\u529f\uf961\u983b\u8b5c\u5bc6\ufa01\u66f2\u7dda\u5716 \u4e09\u3001\u5be6\u9a57\u7d50\u679c\u8207\u5206\u6790\u8a0e\uf941 \u672c\u7ae0\u5c07\u4ecb\u7d39\u672c\uf941\u6587\u76f8\u95dc\u8fa8\uf9fc\u5be6\u9a57\u7684\u5404\uf9d0\u8a2d\u5b9a\uff0c\u7b2c\u4e00\u5c0f\u7bc0\u4ecb\u7d39\u5be6\u9a57\u6240\u7528\u7684 Aurora-2 \u8a9e\u97f3 \u8cc7\uf9be\u5eab\u8207\u8fa8\uf9fc\u6548\u80fd\u8a55\u4f30\u65b9\u5f0f\uff0c\u7b2c\u4e8c\u5c0f\u7bc0\u4ecb\u7d39\u8a9e\u97f3\u8fa8\uf9fc\u5be6\u9a57\u6240\u4f7f\u7528\u7684\u8a9e\u97f3\u8072\u5b78\u6a21\u578b\u3001\u5448\u73fe \u57fa\u672c\u5be6\u9a57\u7684\u8fa8\uf9fc\u7d50\u679c\u4e26\u52a0\u4ee5\u8a0e\uf941\u3002 (\u4e00) \u5be6\u9a57\u74b0\u5883\u8207\u67b6\u69cb\u8a2d\u5b9a \u6211\u5011\u5be6\u9a57\u4e2d\u6240\u63a1\u7528\u7684\u8a9e\u97f3\u8cc7\uf9be\u5eab\u70ba\u6b50\u6d32\u96fb\u4fe1\u6a19\u6e96\u5354\u6703(European Telecommunication Standard Institute, ETSI)\u6240\u767c\ufa08\u7684 Aurora-2 \u8a9e\u97f3\u8cc7\uf9be\u5eab[12] \uff0c\u5b83\u662f\u7531\u7f8e\u570b\u6210\uf98e\u7537\uf981\u4ee5\u4eba\u5de5 \u65b9\u5f0f\uf93f\u88fd\u7684\u4e00\u7cfb\uf99c\uf99a\u7e8c\u82f1\u6587\uf969\u5b57\u5b57\uf905\uff0c\u5176\u4e2d\u6e2c\u8a66\u8a9e\uf9be\u5eab\u7684\u6bcf\u4e00\u5b57\uf905\u4e2d\uff0c\u52a0\u5165\u5404\u7a2e\u52a0\u6210\u6027 \u96dc\u8a0a\u53ca\u901a\u9053\u6548\u61c9\u7684\u5e72\u64fe\uff0c\u9019\u516b\u7a2e\u52a0\u6210\u6027\u7684\u96dc\u8a0a\uff0c\u5206\u5225\u70ba\uff1a\u5730\u4e0b\u9435(subway)\u3001\u4eba\u7684\u5608\u96dc\u8072 (babble)\u3001\u6c7d\uf902(car)\u3001\u5c55\u89bd\u9928(exhibition)\u3001\u9910\u5ef3(restaurant)\u3001\u8857\u9053(street)\u3001\u6a5f\u5834(airport)\u3001 \u706b\uf902\u7ad9(train station)\u7b49\u74b0\u5883\u7684\u96dc\u8a0a\uff0c\u4e26\u4ee5\uf967\u540c\u7a0b\ufa01\u7684\u8a0a\u96dc\u6bd4(signal-to-noise ratio, SNR)\u647b \u96dc\uff0c\u5206\u5225\u70ba\uff1aclean\u300120 dB\u300115 dB\u300110 dB\u30015 dB\u30010 dB \u8207-5 dB\uff1b\u800c\u901a\u9053\u6548\u61c9\u5206\u5225\u70ba.712 \u8207 MIRS \uf978\u7a2e\u901a\u9053\u6a19\u6e96\uff0c\u5b83\u5011\u662f\u900f\u904e\u570b\u969b\u96fb\u4fe1\uf997\u76df(International Telecommunication Union,ITU)[13]\u6240\u8a02\u5b9a\u800c\u6210\u7684\u3002 \u5728\u672c\uf941\u6587\u7576\u4e2d\u4e4b\u6240\u6709\u7684\u8a9e\u97f3\u8fa8\uf9fc\u5be6\u9a57\uff0c\u7686\u4f7f\u7528 12 \u7dad\u6885\u723e\u5012\u983b\u8b5c\u4fc2\uf969(c 1~c12 )\u8207 1 \u7dad\u5c0d\u80fd \uf97e(log-energy)\uff0c\u9644\u52a0\u5176\u4e00\u968e\u5dee\uf97e\u8207\u4e8c\u968e\u5dee\uf97e\uff0c\u5171 39 \u7dad\uff0c\u4f5c\u70ba\u539f\u59cb\u7279\u5fb5\uf96b\uf969\u3002\u800c\u5be6\u9a57\u4e2d \u6240 \u4f7f \u7528 \u7684 \u8072 \u5b78 \u6a21 \u578b (acoustic models) \u662f \u96b1 \u85cf \u5f0f \u99ac \u53ef \u592b \u6a21 \u578b (hidden Markov model, HMM)\u3002\u5728\u8a13\uf996\u65b9\u5f0f\u4e0a\uff0c\u85c9\u7531 HTK[14]\u9019\u5957\u8edf\u9ad4\uf92d\u8a13\uf996\u96b1\u85cf\u5f0f\u99ac\u53ef\u592b\u6a21\u578b\uff0c\u5305\u542b\uf9ba 11 \u500b\uf969\u5b57\u6a21\u578b(zero, one,. . . nine and oh)\u4ee5\u53ca\u4e00\u500b\u975c\u97f3(silence)\u6a21\u578b\uff0c\u6bcf\u500b\uf969\u5b57\u7684 HMM \u5167 \u7686\u5305\u542b 16 \u500b\uf9fa\u614b\uff0c\u800c\u6bcf\u500b\uf9fa\u614b\u662f\u7531 20 \u500b\u9ad8\u65af\u5bc6\ufa01\u51fd\uf969\u7d44\u6210\u3002 (\u4e8c) \u5be6\u9a57\u7d50\u679c\u5448\u73fe\u8207\u8a0e\uf941 \u9996\u5148\uff0c\u6211\u5011\u5728\u8868\u4e00\u4e2d\uff0c\u5448\u73fe\uf9ba\u5e7e\u7a2e\u5728\u7b2c\u4e00\u7ae0\u4e2d\u6240\u63d0\u5230\u7684\u8a9e\u97f3\u7279\u5fb5\u6642\u9593\u5e8f\uf99c\u8655\uf9e4\u6280\u8853\uff0c\u5728 \u4e0a\u8ff0\u5be6\u9a57\u74b0\u5883\u6240\u5f97\u4e4b\u8fa8\uf9fc\u7cbe\u78ba\ufa01\uff0c\u800c\u8868\u4e2d\u7684 AR \u8207 RR \u5206\u5225\u4ee3\u8868\uf9ba\u76f8\u5c0d\u65bc\u57fa\u790e\u5be6\u9a57 (baseline)\u7684\u7d55\u5c0d\u932f\u8aa4\ufa09\u4f4e\uf961\u8207\u76f8\u5c0d\u932f\u8aa4\ufa09\u4f4e\uf961\u3002\u7531\u6b64\u8868\u4e2d\u53ef\u4ee5\u767c\u73fe\uff0c\u9019\u5e7e\u7a2e\u6642\u9593\u5e8f\uf99c\u7279 \u5fb5\u5f37\u5065\u6027\u65b9\u6cd5\uff0c\u90fd\u80fd\u6709\u6548\u6539\u5584\u96dc\u8a0a\u74b0\u5883\u4e0b\u7684\u8fa8\uf9fc\u7d50\u679c\uff0c\u5728\u9019\u4e9b\u65b9\u6cd5\u4e2d\uff0c\u6211\u5011\u4f9d\u7167\u5176\u8fa8\uf9fc \u6548\u80fd\u7531\u9ad8\u81f3\u4f4e\u4f9d\u5e8f\u70ba HEQ\uff0cMVA\uff0cMVMF\uff0cLSSF\uff0cMSI\uff0cCMVN \u8207 CMN\uff0c\u56e0\u6b64\uff0cHEQ \u6cd5\u662f\u6700\u80fd\u6709\u6548\u6539\u5584\u96dc\u8a0a\u74b0\u5883\u5f71\u97ff\u4e0b\u4e4b\u8a9e\u97f3\u8fa8\uf9fc\uff0c\u96d6\u7136\u5176\u4ed6\uf9d1\u7a2e\u65b9\u6cd5\u7684\u8fa8\uf9fc\u6548\u80fd\uf967\u53ca HEQ \u6cd5\uff0c\u4f46\u662f\u76f8\u8f03\u65bc\u57fa\u790e\u5be6\u9a57\u800c\u8a00\uff0c\u5b83\u5011\u90fd\u5c0d\u8a9e\u97f3\u8fa8\uf9fc\u6548\u80fd\u90fd\u6709\u986f\u8457\u7684\u63d0\u5347\uff0c\u4e14\u9664\uf9ba CMN \u8207 CMVN \u5916\uff0c\u5176\u4ed6\u65b9\u6cd5\u8207 HEQ \u6240\u5f97\u4e4b\u8fa8\uf9fc\uf961\u7684\u5dee\u8ddd\u7686\u5728 2%\u4ee5\u5167\u3002 \u8868\u4e00\uff1a\uf969\u7a2e\u8a9e\u97f3\u7279\u5fb5\u6642\u9593\u5e8f\uf99c\u8655\uf9e4\u6280\u8853\u6240\u5f97\u4e4b\u8fa8\uf9fc\u7cbe\u78ba\uf961(%) \u8868\u4e8c\uff1a\uf969\u7a2e\u8a9e\u97f3\u7279\u5fb5\u6642\u9593\u5e8f\uf99c\u8655\uf9e4\u6280\u8853\u6240\u5f97\u4e4b\u8fa8\uf9fc\u7cbe\u78ba\uf961(%) \u5728\u8868\u4e8c\u4e2d\uff0c\u6211\u5011\u6709\u4ee5\u4e0b\u7684\u89c0\u5bdf\u7d50\u679c\uff0c 1. MV-\u6700\u5c0f\u8b8a\uf962\uf969\u7684\u6700\u4f73\u6e96\u5247\uff0c\u5176\u76ee\u6a19\u529f\uf961\u983b\u8b5c\u4e4b\u6c42\u53d6\u4e0a\u540c\u6642\u8003\u616e\uf9ba\u7576\u4e0b\u8655\uf9e4\u4e4b\u55ae\u4e00\u8a9e\uf906\u7684\u529f \uf961\u983b\u8b5c(\u5f0f(15)\u4e2d\u7684 ( ) X k P w )\u8207\u5e73\u5747\u4e7e\u6de8\u529f\uf961\u983b\u8b5c(\u5f0f(15)\u4e2d\u7684 ( ) S k P w ) \uff0c\u5c0d\u8a9e\u97f3\u7279\u5fb5\u53ef \u9054\u5230\u8f03\u4f73\u7684\u5f37\u5065\u5316\u6548\u679c\u3002 \u8868\u4e09\uff1a\uf969\u7a2e\u8a9e\u97f3\u7279\u5fb5\u6642\u9593\u5e8f\uf99c\u8655\uf9e4\u6280\u8853\u6240\u5f97\u4e4b\u8fa8\uf9fc\u7cbe\u78ba\uf961(%) \u56db\u3001\u7d50\uf941\u8207\u672a\uf92d\u5c55\u671b \u672c\uf941\u6587\u4e2d\uff0c\u6211\u5011\u57fa\u65bc\u6700\u5c0f\u8b8a\uf962\uf969\u6e96\u5247\u4e2d\u6240\u5b9a\u7fa9\u7684\u74b0\u5883\u5931\u771f\uff0c\u4f7f\u5176\ufa09\u81f3\u6975\u5c0f\u503c\u9032\u800c\u6c42\u5f97\u983b \u8b5c\u4e0a\u8abf\u8b8a\uf984\u6ce2\u5668\u4e4b\u6700\u4f73\u983b\uf961\u97ff\u61c9\uff0c\u4e26\u61c9\u7528\u65bc\uf978\u7a2e\u8abf\u8b8a\u983b\u8b5c\u6b63\u898f\u5316\u6cd5\uff1a\u6700\u5c0f\u5e73\u65b9\u983b\u8b5c\u64ec\u5408 \u6cd5 \u5728\u672a\uf92d\u5c55\u671b\u4e2d\uff0c\u6211\u5011\u5c07\u9032\u4e00\u6b65\u7814\u7a76\u6700\u5c0f\u8b8a\uf962\uf969\u8abf\u8b8a\uf984\u6ce2\u5668\u6cd5\u7684\uf9e4\uf941\u57fa\u790e\uff0c\u4e26\u5e0c\u671b\u80fd\u85c9\u7531 \uf901\u56b4\u8b39\u7684\uf969\u5b78\u5206\u6790\u8207\u63a8\u5c0e\uff0c\u5c07\u6240\u6c42\u5f97\u7684\u6700\u4f73\u983b\uf961\u97ff\u61c9\u61c9\u7528\u65bc\u5176\u4ed6\u8abf\u8b8a\u983b\u8b5c\u6b63\u898f\u5316\u6cd5\u4e2d\uff0c \u4f7f\u8fa8\uf9fc\u6548\u80fd\u53ef\u4ee5\uf901\u52a0\u63d0\u5347\u3002\u6b64\u5916\uff0c\u6211\u5011\u4e5f\u5e0c\u671b\u76f8\u95dc\u5be6\u9a57\uf967\u50c5\u5728\uf969\u5b57\u8fa8\uf9fc\u4e0a\u8655\uf9e4\uff0c\u4e5f\u64f4\u5c55 \u63a5\u8457\uff0c\u5728\u8868\u4e8c\u4e2d\uff0c\u6211\u5011\u5448\u73fe\u6240\u63d0\u51fa\u7684\uf978\u500b\u65b0\u65b9\u6cd5\uff0cMV-LSSF \u8207 \u51fa\uf901\u65b0\u76ee\u6a19\u8abf\u8b8a\u983b\u8b5c\u529f\uf961\u5bc6\ufa01\u6240\u5e36\uf92d\u7684\u5dee\uf962\u6027\u3002 \u81f3\u5176\u4ed6\u8f03\u5927\u5b57\u5f59\uf97e\u7684\u8a9e\u97f3\u8fa8\uf9fc\uff0c\u6216\u662f\u61c9\u7528\u65bc\u5176\u4ed6\uf9d0\u578b\u7684\u5e72\u64fe\u5931\u771f\u74b0\u5883\uff0c\u63a2\u8a0e\u9019\u4e00\u7cfb\uf99c\u8abf \u8b8a\u983b\u8b5c\u6b63\u898f\u5316\u6cd5\u5728\uf967\u540c\uf9d0\u578b\u4e4b\u8a9e\u97f3\u8fa8\uf9fc\u7cfb\u7d71\u7684\u6548\u80fd\uff0c\u9032\u4e00\u6b65\u9a57\u8b49\u6211\u5011\u63d0\u51fa\u7684\u6539\u9032\u9ede\u8207\u63a2 \u8a0e\u5176\u5be6\u7528\u6027\u3002 2. MV-MSI\u3002 \uf96b\u8003\u6587\u737b</td></tr></table>", |
| "html": null, |
| "type_str": "table", |
| "num": null, |
| "text": "MV-MSI\uff0c\u5728\u4e0a\u8ff0\u5be6\u9a57 \u74b0\u5883\u6240\u5f97\u4e4b\u8fa8\uf9fc\u7cbe\u78ba\uf961\uff0c\u4e26\u5728\u9019\u8868\u4e2d\uff0c\uf99c\u51fa\uf9ba MVMF \u6cd5\u8207\u539f\u59cb LSSF \u8207 MSI \u6cd5\u6240\u5f97 \u4e4b\u8fa8\uf9fc\uf961\uff0c\u4ee5\u4f9b\u6bd4\u8f03\u3002\u5728 MV-LSSF \u8207 MV-MSI \u4e2d\uff0c\u5f0f(15)\u4e2d\u7684\u6bd4\uf9b5\uf96b\uf969l \u8a2d\u5b9a\u70ba 0.5 LSSF \u6cd5\u8207\u539f\u59cb LSSF \u6cd5\u76f8\u8f03\u4e0b\uff0c\u6574\u9ad4\u5e73\u5747\u8fa8\uf9fc\uf961\u53ef\u63d0\u5347 1.13%\uff0c\u800c MV-MSI \u6cd5 \u76f8\u8f03\u65bc\u539f\u59cb MSI \u6cd5\u800c\u8a00\uff0c\u6574\u9ad4\u5e73\u5747\u8fa8\uf9fc\uf961\u53ef\u63d0\u5347 0.39%\u3002\u7531\u6b64\u53ef\u77e5\uff0cMV-LSSF \u8207 MV-MSI \u5206\u5225\u512a\u65bc\u539f\u59cb\u4e4b LSSF \u8207 MSI\uff0c\u53ef\u80fd\u539f\u56e0\u70ba\uff0cMV-LSSF \u8207 MV-MSI \u4f7f\u7528\uf9ba \u5728\u8abf\u8b8a\u983b\u8b5c\u57df\u63a8\u5f97\u7684 MV-LSSF \u6cd5\u548c MV-MSI \u6cd5\uff0c\u8207\u6642\u9593\u5e8f\uf99c\u57df\u63a8\u5f97\u7684 MVMF \u6cd5 \u76f8\u6bd4\u8f03\uff0c\u5728\u6574\u9ad4\u7684\u5e73\u5747\u8fa8\uf9fc\uf961\u4e0a\uff0cMVMF-LSSF \u6cd5\u7d04\u6709 1%\u7684\u9032\u6b65\uf961\uff0c\u800c MVMF-MSI \u5247\u7d04\u8207 MVMF \u6cd5\u76f8\u7b49(\u4e9b\u5fae\u9000\u6b65\uf9ba 0.15%) \u3002\u7136\u800c\uff0c\u7531\u65bc MVMF \u5fc5\u9808\u4f7f\u7528\u5230\u53cd\u77e9\u9663\u7684 \u904b\u7b97(\u5982\u5f0f(7)\u6240\u793a) \uff0c\u76f8\u5c0d\u65bc MV-LSSF \u8207 MV-MSI \u800c\u8a00\uff0c\u57f7\ufa08\u4e0a\u8907\u96dc\ufa01\u8f03\u9ad8\uff0c\u7531\u6b64\u770b \u51fa\u6211\u5011\u6240\u65b0\u63d0\u51fa\u7684\uf978\u7a2e\u65b9\u6cd5\uff0c\u65e2\u53ef\u8207 MVMF \u6cd5\u6548\u679c\u4e26\u99d5\u9f4a\u9a45\u6216\uf976\u4f73\uff0c\u4e14\u5728\u904b\u7b97\u4e0a\uf901\u6709 \u6548\uf961\uff0c\u56e0\u6b64\uf901\u5177\u512a\u52e2\u3002\u540c\u6642\uff0c\u6211\u5011\u6240\u63d0\u51fa\u7684 MV-LSSF \u6cd5\uff0c\u5176\u9054\u5230\u7684\u8fa8\uf9fc\uf961\u9032\u6b65\u7a0b\ufa01\u5df2 \u7d93\u8207\u8868\u4e00\u6240\u8a0e\uf941\u4e4b\u6700\u4f73\u6548\u679c\u7684 HEQ \u6cd5\u5e7e\u4e4e\u76f8\u540c\uff0c\u8db3\ufa0a\u5176\u512a\u8d8a\u6027\u3002 \u6700\u5f8c\uff0c\u6211\u5011\u5617\u8a66\u5c07\u6240\u63d0\u51fa\u4e4b MV-LSSF \u8207 MV-MSI \u6cd5\uff0c\u8207\u7279\u5fb5\u7d71\u8a08\u6b63\u898f\u5316\u6cd5\u4e4b\u4e00\u7684 CMVN \u6cd5\u7d50\u5408\uff0c\u89c0\u5bdf\u5176\u662f\u5426\u80fd\u4fc3\u6210\u8fa8\uf9fc\uf961\uf901\u9032\u4e00\u6b65\u7684\u63d0\u5347\uff0c\u5728\u6b64\uff0c\u539f\u59cb\u8a9e\u97f3\u5012\u983b\u8b5c\u7279 \u5fb5\u5148\u7d93\u904e CMVN \u8655\uf9e4\u5f8c\uff0c\u518d\u5206\u5225\u7d93\u904e\u4e09\u7a2e MV \u6cd5\uff0c\u5373 MVMF\u3001 MV-LSSF \u8207 MV-MSI\uff0c \u6240\u5f97\u4e4b\u8fa8\uf9fc\uf961\u8a73\uf99c\u65bc\u8868\u4e09\u3002\u5f9e\u6b64\u8868\u4e2d\uff0c\u6211\u5011\u660e\u986f\u770b\u51fa\uff0c\u4e0a\u8ff0\u4e09\u7a2e MV \u6cd5\u8207 CMVN \u6cd5\u7686 \u6709\u660e\u986f\u7684\u52a0\u6210\u6027\uff0c\u76f8\u5c0d\u65bc\u55ae\u4e00 CMVN \u6cd5\u800c\u8a00\uff0c\u5c07 MVMF\u3001 MV-LSSF \u8207 MV-MSI \u8655 \uf9e4\u65bc CMVN \u5f8c\u7684\u7279\u5fb5\u4e0a\uff0c\u5206\u5225\u6709 6.66%\u30017.24%\u8207 6.50%\u4e4b\u5e73\u5747\u8fa8\uf9fc\uf961\u7684\u63d0\u5347\uff0c\uf974\u5c07\u8868 \u4e09\u8207\u8868\u4e8c\u76f8\u8f03\uff0c\u4e5f\u53ef\u770b\u51fa\uff0c\u7d50\u5408 CMVN \u6cd5\u5f8c\uff0c\u4e09\u7a2e MV \u6cd5\u4e5f\u80fd\u6709\uf901\u660e\u986f\u9032\u6b65\u7684\u6548\u80fd\u3002 \u8ddf\u4e4b\u524d\u7d50\u679c\uf9d0\u4f3c\uff0c\u7d50\u5408\uf9ba CMVN \u6cd5\u5f8c\uff0cMV-LSSF \u7684\u6548\u679c\u4ecd\u7136\u6700\u597d\uff0c\u5176\u6b21\u70ba MVMF \u8207 (least-squares spectrum fitting, LSSF) \u8207 \u5f37 \ufa01 \u983b \u8b5c \u5167 \u63d2 \u6cd5 (magnitude spectrum interpolation, MSI)\uff0c\u9032\u800c\u767c\u5c55\u51fa\uf9ba\u65b0\u7684\uf978\u7a2e\u65b0\u65b9\u6cd5\uff0c\u5373 MV-LSSF \u8207 MV-MSI\u3002\u7531\u5be6\u9a57 \u7d50\u679c\u767c\u73fe\uff0c\u5728\u8a9e\u97f3\u8fa8\uf9fc\u6548\u80fd\u4e0a\uff0c\u9019\uf978\u7a2e\u5728\u8abf\u8b8a\u983b\u8b5c\u57df\u6240\u767c\u5c55\u7684\u65b0\u65b9\u6cd5\uff0c\uf974\u8207\u6642\u9593\u5e8f\uf99c\u57df \u7684 MVMF \u6cd5\u76f8\u6bd4\u8f03\uff0cMV-LSSF \u6cd5\u8fa8\uf9fc\uf961\u9032\u6b65\u7d04\u6709 1.13%\uff0c\u800c MV-MSI \u5247\u7d04\u8207 MVMF \u6cd5\u76f8\u7b49\uff0c\u4f46 MV-LSSF \u548c MV-LSI \u6bd4 MVMF \u6709\u8f03\u4f4e\u7684\u904b\u7b97\u8907\u96dc\ufa01\u3002" |
| } |
| } |
| } |
| } |
