| { |
| "paper_id": "O09-2005", |
| "header": { |
| "generated_with": "S2ORC 1.0.0", |
| "date_generated": "2023-01-19T08:11:19.424228Z" |
| }, |
| "title": "Speech Enhancement Technique Based on Blind Source Separation for Far-Field Noisy Speech Recognition", |
| "authors": [ |
| { |
| "first": "Sheng-Chieh", |
| "middle": [], |
| "last": "\uf9e1\u8056\u6377", |
| "suffix": "", |
| "affiliation": { |
| "laboratory": "", |
| "institution": "Cheng Kung University", |
| "location": {} |
| }, |
| "email": "" |
| }, |
| { |
| "first": "", |
| "middle": [], |
| "last": "Lee", |
| "suffix": "", |
| "affiliation": { |
| "laboratory": "", |
| "institution": "Cheng Kung University", |
| "location": {} |
| }, |
| "email": "" |
| }, |
| { |
| "first": "Jhing-Fa", |
| "middle": [], |
| "last": "\u738b\u99ff\u767c", |
| "suffix": "", |
| "affiliation": { |
| "laboratory": "", |
| "institution": "Cheng Kung University", |
| "location": {} |
| }, |
| "email": "" |
| }, |
| { |
| "first": "", |
| "middle": [], |
| "last": "Wang", |
| "suffix": "", |
| "affiliation": { |
| "laboratory": "", |
| "institution": "Cheng Kung University", |
| "location": {} |
| }, |
| "email": "wangjf@csie.ncku.edu.tw" |
| }, |
| { |
| "first": "Miao-Hai", |
| "middle": [], |
| "last": "\u9673\u6dfc\u6d77", |
| "suffix": "", |
| "affiliation": { |
| "laboratory": "", |
| "institution": "Cheng Kung University", |
| "location": {} |
| }, |
| "email": "" |
| }, |
| { |
| "first": "", |
| "middle": [], |
| "last": "Chen", |
| "suffix": "", |
| "affiliation": { |
| "laboratory": "", |
| "institution": "Cheng Kung University", |
| "location": {} |
| }, |
| "email": "" |
| } |
| ], |
| "year": "", |
| "venue": null, |
| "identifiers": {}, |
| "abstract": "Speech recognition is one of the im portant p arts of search field in speech processing. Nevertheless, the speec h environment and speec h distance w ill mainly affect the reco gnition result. In this paper, a high adaptation far-fie ld noise speech recognition system is proposed. This system is combined with the m ethods of independent component analysis and subspace speech enh ancement, and then furth er filteri ng the noise of speech to improve the speech quality for recognition. The experimental results show that the proposed system is suitable for several presented noisy environments, and it can effectively improve the recognition rate. For the SNR evaluation, this proposed sy stem can make enhanced speech SNR with 20dB higher than original corrupted speech which ranges from 0dB to 10dB.", |
| "pdf_parse": { |
| "paper_id": "O09-2005", |
| "_pdf_hash": "", |
| "abstract": [ |
| { |
| "text": "Speech recognition is one of the im portant p arts of search field in speech processing. Nevertheless, the speec h environment and speec h distance w ill mainly affect the reco gnition result. In this paper, a high adaptation far-fie ld noise speech recognition system is proposed. This system is combined with the m ethods of independent component analysis and subspace speech enh ancement, and then furth er filteri ng the noise of speech to improve the speech quality for recognition. The experimental results show that the proposed system is suitable for several presented noisy environments, and it can effectively improve the recognition rate. For the SNR evaluation, this proposed sy stem can make enhanced speech SNR with 20dB higher than original corrupted speech which ranges from 0dB to 10dB.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Abstract", |
| "sec_num": null |
| } |
| ], |
| "body_text": [ |
| { |
| "text": "EQUATION", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [ |
| { |
| "start": 0, |
| "end": 8, |
| "text": "EQUATION", |
| "ref_id": "EQREF", |
| "raw_str": "\uf0fa \uf0fb \uf0f9 \uf0ea \uf0eb \uf0e9 \uf03d 22 21 12 11 a a a a A \u56e0\u6b64\uf974\u5047\u8a2d\u8072\u6e90\u8a0a\u865f s 1 \u70ba\u8a9e\u8005\u7684\u8a9e\u97f3\u6210\u5206\u3001s 2 \u70ba\u566a\u97f3\u6210\u5206\uff0c\u6211\u5011\u53ef\u5f9e\u6240\u63a5\u6536\u5230\u4e4b\u6df7\u5408\u8a0a \u865f x 1 \u53ca x 2 \u5206\uf9ea\u51fa\u539f\u59cb\u7684\u8a9e\u97f3\u8a0a\u865f\u4ee5\u53ca\u566a\u97f3\u8a0a\u865f\uff0c\u5373\u53ef\u6709\u6548\u7684\u53bb\u9664\u96dc\u8a0a\uff0c\u6839\u64da\u4e0a\u8ff0\u516c\u5f0f (1)\uff0c\u8981\u6c42\u5f97\u539f\u59cb\u8a0a\u865f x 1 \u548c x 2 \uff0c\u5fc5\u9808\u627e\u51fa\u4e00\u500b\u89e3\u6df7\u5408\u77e9\u9663 A \uff0c\u4f7f\u5f97\u63a5\u6536\u5230\u7684\u6df7\u5408\u8a0a\u865f\u7d93 \u7531 A \u8f49\u63db\u5f8c\uff0c\u53ef\u5f97\u5230\u539f\uf92d\u7684\u8072\u6e90\u8a0a\u865f\uff0c\u800c\u6b64\u6c42\u89e3 A \u4e4b\u65b9\u6cd5\u5373\u70ba\u7368\uf9f7\u6210\u5206\u5206\u6790\u6cd5\u3002 \u5728\u4f7f\u7528\u7368\uf9f7\u6210\u5206\u5206\u6790\u6cd5\u6c42\u5f97\u89e3\u6df7\u5408\u77e9\u9663\u524d\uff0c\u5fc5\u9808\u5148\ufa08\u5047\u8a2d\u8a0a\u865f\u6e90\u5f7c\u6b64\u7368\uf9f7\uff0c\u7136\u800c\u5728\u771f\u5be6 \u60c5\u6cc1\u4e0b\uff0c\u8a0a\u865f\u6e90\u4e26\u975e\u90fd\u6703\u5f7c\u6b64\u4e92\u76f8\u7368\uf9f7\uff0c\u56e0\u6b64\u5728\u9032\ufa08\u7368\uf9f7\u6210\u5206\u5206\u6790\uf9ca\u7a0b\u524d\uff0c\u5fc5\u9808\u5148\u7d93\u904e \u524d\u7f6e\u8655\uf9e4\u5f8c\u624d\u80fd\u627e\u5c0b\u89e3\u6df7\u5408\u77e9\u9663\uff0c\u5728\u6b64\u6211\u5011\u524d\u7f6e\u8655\uf9e4\u65b9\u5f0f\u70ba\u96c6\u4e2d\u8b8a\uf969(Centering)\u4ee5\u53ca\u8cc7 \uf9be\u767d\u8272\u5316(Whitening)\u8655\uf9e4\uff0c\u5728\u6b64\u5148\u91dd\u5c0d\u96c6\u4e2d\u8b8a\uf969\u53ca\u8cc7\uf9be\u767d\u8272\u5316\uf92d\u4f5c\u70ba\uf96f\u660e\u3002 1 \u96c6\u4e2d\u8b8a\uf969(Centering) \u96c6\u4e2d\u8b8a\uf969\u7684\u8655\uf9e4\u6b65\u9a5f\u4e3b\u8981\u662f\u5c07\u6df7\u5408\u8a0a\u865f\u6263\u9664\u5176\u5e73\u5747\u503c\uff0c\u85c9\u6b64\u7c21\u5316\u4e4b\u5f8c\u6c42\u5f97\u89e3\u6df7\u5408\u77e9\u9663\u4e4b \u6c42\u89e3\u904e\u7a0b\uff0c\u5176\u516c\u5f0f\u5982\u4e0b\u6240\u8868\u793a\u3002 } { x E x x \uf02d \uf03d (2) \u6b64\u5916\u6211\u5011\u5c07\u63a5\u6536\u5230\u7684\u6df7\u5408\u8a0a\u865f\u505a\u96c6\u4e2d\u8b8a\uf969\u8655\uf9e4\u5f8c\uff0c\u4e5f\u540c\u6a23\u7684\u5c0d\u65bc\u8072\u6e90\u8a0a\u865f\u505a\uf9ba\u96c6\u4e2d\u8b8a\uf969 \u8655\uf9e4\uff0c\u5982\u516c\u5f0f(3)\u6240\u793a\u3002 0 } { } { } { \uf03d \uf03d \uf03d x E A x A E s E (3) 2 \u8cc7\uf9be\u767d\u8272\u5316(Whitening) \u81f3\u65bc\u524d\u7f6e\u8655\uf9e4\u7684\u7b2c\u4e8c\u6b65\u9a5f\u5c31\u662f\u8cc7\uf9be\u767d\u8272\u5316\uff0c\u8cc7\uf9be\u767d\u8272\u5316\u7684\u76ee\u7684\u5728\u65bc\u5c07\u8f49\u63db\u5f8c\u7684\u8cc7\uf9be\u5f7c\u6b64 \u9593\u5177\u6709\u975e\u76f8\u95dc\u6027(Uncorrelated)\u4e14\u8b8a\uf962\uf969(Variance)\uf969\u503c\u70ba\u4e00\uff0c\u5728\u6b64\u5047\u8a2d\u8f49\u63db\u8cc7\uf9be\u70ba z\uff0c\u5247 \u6b64\u8cc7\uf9be\u4e4b\u5171\u8b8a\uf962\u77e9\u9663(Covariance matrix)\u6703\u6210\u70ba\u55ae\u4f4d\u77e9\u9663\u3002\u56e0\u6b64\u8cc7\uf9be\u767d\u8272\u5316\u7684\u65b9\u5f0f\u70ba\u627e\u51fa \u4e00\u767d\u8272\u5316\u77e9\u9663 V\uff0c\u4e26\u5c07\u6240\u63a5\u6536\u5230\u7684\u8a0a\u865f x \u505a\u7dda\u6027\u8f49\u63db\u4e14\u4f7f\u5176\u5171\u8b8a\uf962\u77e9\u9663\u70ba\u55ae\u4f4d\u77e9\u9663\u3002 (4) I zz E Vx z T \uf03d \uf03d } { , 3 \u89e3\u6df7\u5408\u77e9\u9663(De-mixing matrix) \u505a\u5b8c\u524d\u7f6e\u8655\uf9e4\u5f8c\uff0c\u518d\uf92d\u5247\u662f\u8a08\u7b97\u6700\u5927\u975e\u9ad8\u65af\u5206\u4f48\u8a0a\u865f\uff0c\u6839\u64da\u4e2d\u592e\u6975\u9650\u5b9a\uf9e4\uff0c\u5c07\u591a\u500b\u975e\u9ad8 \u65af\u5206\u5e03\u4e14\u5f7c\u6b64\u7368\uf9f7\u7684\u8a0a\u865f\u500b\u5225\u52a0\u7e3d\u5f8c\uff0c\u6703\u4f7f\u5f97\u6574\u9ad4\u50be\u5411\u65bc\u9ad8\u65af\u5206\u4f48\uff0c\u56e0\u6b64\uf974\u4efb\uf978\u500b\u96a8\u6a5f \u8a0a\u865f\u8d8a\u50be\u5411\u975e\u9ad8\u65af\u5206\u4f48\uff0c\u5247\u6b64\uf978\u8a0a\u865f\u5f7c\u6b64\u7368\uf9f7\u7684\u6210\u5206\u5c31\u8d8a\u5927\uff0c\u518d\u6b64\u9ad8\u65af\u5206\u4f48\u8a0a\u865f\u5177\u6709\u758a \u52a0\u6027\uff0c\uf978\u500b\u9ad8\u65af\u5206\u4f48\u8a0a\u865f\u76f8\u52a0\u7e3d\u5f8c\u7684\u8a0a\u865f\u4ecd\u70ba\u9ad8\u65af\u5206\u4f48\uff0c\u6240\u4ee5\uf974\u7531\u9ad8\u65af\u8a0a\u865f\u7dda\u6027\u6df7\u5408\u800c \u6210\u7684\u7fa4\u96c6\u662f\u7121\u6cd5\u5206\uf9ea\u51fa\u771f\u6b63\u7684\u539f\u59cb\u8a0a\u865f\uff0c\u56e0\u6b64\u5728\u4f7f\u7528\u7368\uf9f7\u6210\u5206\u5206\u6790\uf92d\u5206\uf9ea\u8a0a\u865f\u6642\uff0c\u5fc5\u9808 \u4e8b\u5148\u5047\u8a2d\u53ea\u80fd\u5141\u8a31\u5176\u4e2d\u4e00\u500b\u8a0a\u865f\u70ba\u9ad8\u65af\u5206\u4f48\uff0c\u5728\u4f30\u7b97\u975e\u9ad8\u65af\u5206\u4f48\u8a0a\u865f\u90e8\u4efd\uff0c\u672c\uf941\u6587\u662f\u63a1 \u7528\u8ca0\u71b5(Negentropy)\uf92d\u8a55\u4f30\u8a08\u7b97\uff0c\u5176\u4e2d\u71b5(Entropy)\u7684\u5b9a\u7fa9\u6839\u64da\uf9ea\u6563\u8a0a\u865f\u6216\uf99a\u7e8c\u8a0a\u865f\u53ef\u7531 \u4e0b\uf99c\u516c\u5f0f\u6240\u8868\u793a\uff1a ) ( log ) ( ) ( y P y P y H \uf0e5 \uf02d \uf03d (5) dy y f y f y H ) ( log ) ( ) ( \uf0f2 \uf02d \uf03d (6) \u5728\u8a9e\u97f3\u8a0a\u865f\u90e8\u4efd\u4e2d\uff0c\u7576\u8a0a\u865f y \u70ba\u9ad8\u65af\u5206\u4f48\u6642\uff0c\u5176\u71b5\u70ba\u6700\u5927\u503c\uff0c\u56e0\u6b64\u70ba\uf9ba\u8a08\u7b97\u65b9\uf965\u6211\u5011\u4f7f \u7528\u8ca0\u71b5\uf92d\u4f5c\u70ba\u4f9d\u64da\uff0c\u5982\u516c\u5f0f(7)\u6240\u793a\uff0cy gauss \u70ba\u548c y \u6709\u76f8\u540c\u8b8a\uf962\u77e9\u9663\u4e4b\u9ad8\u65af\u5206\u4f48\u8a0a\u865f\uff0c\u56e0 \u6b64\u7576\u8a0a\u865f y \u70ba\u9ad8\u65af\u5206\u4f48\u6642\uff0c\u5247\u8ca0\u71b5\u70ba\uf9b2\uff0c\u70ba\uf9ba\u7c21\u5316\u5176\u8a08\u7b97\uff0c\u6211\u5011\u5c07\u516c\u5f0f(7)\u7c21\u5316\u70ba\u516c\u5f0f(8)\u3002 ) ( ) ( ) ( y H y H y J gauss \uf02d \uf03d (7) (8) 2 )}] ( { )} ( { [ ) ( v G E y G E y J \uf02d \uf0bb \u5176\u4e2d G \u70ba\u5c0d\u7167\u65b9\u7a0b\u5f0f\uff0c\u8a0a\u865f v \u70ba\u5e73\u5747\u503c\u70ba\uf9b2\u8b8a\uf962\uf969\u70ba\u4e00\u4e4b\u9ad8\u65af\u5206\u4f48\u8a0a\u865f\uff0c\u4e00\u822c\uf92d\uf96f\u5c0d \u7167\u65b9\u7a0b\u5f0f\uf967\u80fd\u70ba\u4e8c\u6b21\u5f0f\u51fd\uf969\u6216\u591a\u9805\u5f0f\u51fd\uf969\uff0c\u5728\u6b64\u6211\u5011\u9078\u64c7\u7684\u5c0d\u7167\u65b9\u7a0b\u5f0f\u5982\u4e0b\u6240\u793a\u3002 \u70ba\u4e00\u5e38\uf969 1 1 1 1 , )) log(cosh( 1 ) ( a y a a y G \uf03d (9) ) 2 exp( ) ( 2 2 y y G \uf02d \uf02d \uf03d (10) (11) 4 3 ) ( y y G \uf03d \u6839\u64da\u4e0a\u9762\u5c0d\u7167\u65b9\u7a0b\u5f0f\uff0c\u6211\u5011\u8a2d\u5b9a E{G(y)}=E{G(W T x)}\uff0cW \u70ba\u89e3\u6df7\u5408\u77e9\u9663\uff0cx \u70ba\u6df7\u5408\u8a0a\u865f\uff0c \u56e0\u6b64\u516c\u5f0f(8)\u53ef\u6539\u5beb\u6210\u516c\u5f0f(12)\uff0c\u7576 E{G(W T x)}\u70ba\u6700\u5927\u6642\uff0c\u5247\u53ef\u627e\u5230\u975e\u9ad8\u65af\u5206\u4f48\u6027\u6700\u9ad8\u7684 \u8a9e\u97f3\u8a0a\u865f\uff0c\u6700\u5f8c\u518d\uf9dd\u7528\u725b\u9813\u6cd5\u758a\u4ee3\u904b\u7b97\uff0c\u5c07\u89e3\u6df7\u5408\u77e9\u9663 W \u6c42\u89e3\u51fa\uf92d\u3002 (12) 2 )}] ( { )} ( { [ ) ( v G E x W G E W J T \uf02d \uf0b5 (13) W x W G E x W xG E W T T )} ( ' { )} ( { \uf02d \uf0ac (\u4e8c)\u5b50\u7a7a\u9593\u8a9e\u97f3\u589e\u5f37\u6cd5(Subspace Speech Enhancement) \u7d93\u7531\u7368\uf9f7\u6210\u5206\u5206\u6790\u6cd5\uff0c\u6211\u5011\u53ef\u5c07\u6df7\u5408\u8a0a\u865f\u5206\uf9ea\u51fa\uf978\u500b\u8a0a\u865f\uff0c\u5176\u4e2d\u4e00\u500b\u8a0a\u865f\u5176\u8a9e\u97f3\u6210\u5206\u8f03 \u5927\uff0c\u53e6\u4e00\u500b\u5247\u662f\u96dc\u8a0a\u6210\u5206\u8f03\u5927\uff0c\u7136\u800c\u542b\u8a9e\u97f3\u6210\u5206\u8f03\u591a\u7684\u8a0a\u865f\u4e2d\uff0c\u4ecd\u820a\u6703\u6b98\uf9cd\u4e9b\u8a31\u96dc\u8a0a\u90e8 \u4efd\uff0c\u56e0\u6b64\u6211\u5011\u4f7f\u7528\u5b50\u7a7a\u9593\u8a9e\u97f3\u589e\u5f37\u6cd5\uf92d\u9032\u4e00\u6b65\u52a0\u5f37\u8655\uf9e4\uff0c\uf984\u9664\u8a0a\u865f\u4e2d\u7684\u566a\u8072\u96dc\u8a0a\u3002 \u5728\u8a0a\u865f\u5b50\u7a7a\u9593\u7684\u5047\u8a2d\u4e2d\uff0c\u53ef\u5c07\u89c0\u6e2c\u8a0a\u865f\u7684\u5411\uf97e\u62c6\u89e3\u6210\uf978\u500b\u5b50\u7a7a\u9593\uff0c\u4e00\u500b\u70ba\u7531\u4e7e\u6de8\u8a9e\u97f3\u8a0a \u865f\u7d44\u5408\u800c\u6210\u7684\u5b50\u7a7a\u9593\uff0c\u53e6\u4e00\u500b\u662f\u8207\u4e7e\u6de8\u8a9e\u97f3\u7a7a\u9593\u6b63\u4ea4(orthogonal)\u4e14\u7531\u566a\u97f3\u6240\u7d44\u6210\u7684\u5b50\u7a7a \u9593\uff0c\u7531\u65bc\u566a\u97f3\u6240\u7d44\u6210\u7684\u5b50\u7a7a\u9593\u6c92\u6709\u4efb\u4f55\u8a9e\u97f3\u8cc7\u8a0a\u56e0\u6b64\u53ef\u5c07\u6b64\u5ffd\uf976\uff0c\u800c\u4e7e\u6de8\u8a9e\u97f3\u8a0a\u865f\u7684\u5b50 \u7a7a\u9593\u4e2d\uff0c\u4ecd\u820a\u6703\u6709\u566a\u97f3\u6210\u5206\u8207\u5176\u4e26\u5b58\uff0c\uf9b5\u5982\u5404\u983b\u5e36\u7686\u6709\u53ef\u80fd\u5b58\u5728\u7684\u767d\u566a\u97f3(White noise)\uff0c \u56e0\u6b64\u8981\u6839\u64da\u566a\u97f3\u6210\u5206\u7684\u5206\u4f48\uf92d\u8655\uf9e4\uff0c\u9084\u539f\u51fa\u6c92\u6709\u96dc\u8a0a\u8a0a\u865f\u7684\u8a9e\u97f3\u5b50\u7a7a\u9593\u3002 \u6211\u5011\u5047\u8a2d\u8a0a\u865f\u5b50\u7a7a\u9593\u4e2d\u4e7e\u6de8\u7684\u8a9e\u97f3\u6210\u5206\u53ef\u7531\u4e00\u7dda\u6027\u6a21\u578b\u7d44\u6210\uff0c\u5982\u516c\u5f0f(14)\u6240\u793a\uff0c\u5176\u4e2d W S \u70ba\u4e00 N\u00d7M \u4e14 M \u5c0f\u65bc N \u7684\u77e9\u9663\uff0cx S \u70ba M\u00d71 \u7684\u5411\uf97e\uff0c\u5247\u6b64\u8a0a\u865f\u5411\uf97e y \u70ba\u4e00\u500b\u7531 Ws \u6240 \u5efa\uf9f7\u7684\u6b50\u57fa\uf9e9\u5fb7\u7a7a\u9593 R N \uf9e8\u7684\u4e00\u500b\u96c6\u5408\uff0c\u800c\u6b64\u7a7a\u9593\u5c31\u662f\u8a0a\u865f\u5b50\u7a7a\u9593\u3002 S S x W y \uf03d (14) \u56e0\u6b64\u539f\u59cb\u6df7\u5408\u8a0a\u865f\u5373\u70ba\u539f\u672c\u7684\u8a9e\u97f3\u8a0a\u865f\u5b50\u7a7a\u9593 y \u518d\u52a0\u4e0a\u53e6\u4e00\u500b\u566a\u97f3\u8a0a\u865f\u5b50\u7a7a\u9593 nS \uff0c\u5982(15) \u5f0f\u6240\u793a\uff0c\u7531\u65bc\u672c\uf941\u6587\u7684\u65b9\u6cd5\u662f\u63a1\u7528\u5728\u6642\u9593\u57df(Time domain)\u4e0b\u7684\u4f30\u7b97\uff0c\u6240\u4ee5\u5728\u6b64\u5c31\u76f4\u63a5\u63a2 \u8a0e\u5728\u6642\u9593\u57df\u4e0b\u7684\u76f8\u95dc\u4f30\u6e2c\u3002 S S S S n y n x W z \uf02b \uf03d \uf02b \uf03d (15) \u6839\u64da\u4e0a\u5f0f\u7684\u6df7\u5408\u8a0a\u865f\uff0c\u6211\u5011\u5fc5\u9808\u627e\u51fa\u4e00\u500b N\u00d7N \u7684\uf984\u6ce2\u5668 F\uff0c\u4f7f\u5f97\u6df7\u5408\u8a0a\u865f\u7d93\u7531\uf984\u6ce2\u5f8c \u80fd\u5f97\u5230\u4e7e\u6de8\u7684\u8a0a\u865f y'=Fz\uff0c\u800c\uf984\u6ce2\u5f8c\u7684\u8a0a\u865f\u8207\u539f\u8a0a\u865f\u76f8\u6bd4\u8f03\u53ef\u8a08\u7b97\u5176\uf984\u6ce2\u5668 F \u7684\u8aa4\u5dee\uff0c \u5176\u8aa4\u5dee\u503c\u03b4\u8a08\u7b97\u5982\u4e0b\uff1a S n y S Fn y I F y y \uf064 \uf064 \uf064 \uf02b \uf03d \uf02b \uf02d \uf03d \uf02d \uf03d ) ( ' (16) \u5176\u4e2d\u03b4 y \u8868\u793a\u88ab\uf984\u6ce2\u5668\uf984\u9664\u7684\u8a9e\u97f3\u8a0a\u865f\u5931\u771f\uff0c S n \uf064 \u8868\u793a\u6c92\u6709\u88ab\uf984\u9664\u7684\u566a\u97f3\u6240\u7522\u751f\u7684\u5931\u771f\uff0c \u56e0\u6b64\u6211\u5011\u8a08\u7b97\u9019\uf978\u7a2e\u5931\u771f\u8aa4\u5dee\u7684\u8b8a\uf962\uf969\u7576\u6210\u5f37\u5316\u5f8c\u7684\u8aa4\u5dee\u80fd\uf97e\u3002 } { y T y y E \uf064 \uf064 \uf064 \uf03d (17) } { S S S n T n n E \uf064 \uf064 \uf064 \uf03d", |
| "eq_num": "(1)" |
| } |
| ], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "[2] B.N. Gover, J.G. Ryan, and M.R. Stinson, \"Microphone array measurement system for analysis of directional and spatial variations of sound fields,\" J. Acoust. Soc. Am., 112,", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| } |
| ], |
| "back_matter": [], |
| "bib_entries": { |
| "BIBREF0": { |
| "ref_id": "b0", |
| "title": "Measurements of directional properties of reverberant sound fields in rooms using a spherical microphone array", |
| "authors": [ |
| { |
| "first": "B", |
| "middle": [ |
| "N" |
| ], |
| "last": "Gover", |
| "suffix": "" |
| }, |
| { |
| "first": "J", |
| "middle": [ |
| "G" |
| ], |
| "last": "Ryan", |
| "suffix": "" |
| }, |
| { |
| "first": "M", |
| "middle": [ |
| "R" |
| ], |
| "last": "Stinson", |
| "suffix": "" |
| } |
| ], |
| "year": null, |
| "venue": "J. Acoust. Soc. Am", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "B.N. Gover, J.G. Ryan, and M.R. Stinson, \"Measurements of directional properties of reverberant sound fields in rooms using a spherical microphone array,\" J. Acoust. Soc. Am. (in press).", |
| "links": null |
| }, |
| "BIBREF1": { |
| "ref_id": "b1", |
| "title": "An optimum microphone array post-filter for speech applications", |
| "authors": [ |
| { |
| "first": "S", |
| "middle": [], |
| "last": "Leukimmiatis", |
| "suffix": "" |
| }, |
| { |
| "first": "D", |
| "middle": [], |
| "last": "Dimitriadis", |
| "suffix": "" |
| }, |
| { |
| "first": "P", |
| "middle": [], |
| "last": "Maragos", |
| "suffix": "" |
| } |
| ], |
| "year": 2006, |
| "venue": "ICSLP", |
| "volume": "", |
| "issue": "", |
| "pages": "2142--2145", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Leukimmiatis, S., Dimitriadis, D., and Maragos, P, \"An optimum microphone array post-filter for speech applications,\" ICSLP, 2006, pp. 2142-2145.", |
| "links": null |
| }, |
| "BIBREF2": { |
| "ref_id": "b2", |
| "title": "Methods for the blind signal separation problem", |
| "authors": [ |
| { |
| "first": "Yan", |
| "middle": [], |
| "last": "Li", |
| "suffix": "" |
| }, |
| { |
| "first": "P", |
| "middle": [], |
| "last": "Wen", |
| "suffix": "" |
| }, |
| { |
| "first": "D", |
| "middle": [], |
| "last": "Powers", |
| "suffix": "" |
| } |
| ], |
| "year": 2003, |
| "venue": "Proc. IEEE Int. Conf. Neural Network, Signal Processing", |
| "volume": "", |
| "issue": "", |
| "pages": "1386--1389", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Yan Li, P. Wen and D. Powers, \"Methods for the blind signal separation problem,\" in Proc. IEEE Int. Conf. Neural Network, Signal Processing, Nanjing China, Dec. 2003, pp. 1386-1389.", |
| "links": null |
| }, |
| "BIBREF3": { |
| "ref_id": "b3", |
| "title": "Space or time adaptive signal processing by neural network models", |
| "authors": [ |
| { |
| "first": "J", |
| "middle": [], |
| "last": "Herault", |
| "suffix": "" |
| }, |
| { |
| "first": "C", |
| "middle": [], |
| "last": "Lutten", |
| "suffix": "" |
| } |
| ], |
| "year": 1986, |
| "venue": "Neural Nehvorks For Computing: AIP Conference Proceedings 151, American Institute for Physics", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "J. Herault and C. lutten, \"Space or time adaptive signal processing by neural network models,\" In J. S. Denkcr (ed), editor, Neural Nehvorks For Computing: AIP Conference Proceedings 151, American Institute for Physics, New York, 1986.", |
| "links": null |
| }, |
| "BIBREF4": { |
| "ref_id": "b4", |
| "title": "Blind separation of sources ~ a nonlinear neural algorithm", |
| "authors": [ |
| { |
| "first": "G", |
| "middle": [], |
| "last": "Burel", |
| "suffix": "" |
| } |
| ], |
| "year": 1992, |
| "venue": "Neural Nehvorkr", |
| "volume": "5", |
| "issue": "6", |
| "pages": "937--947", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "G. Burel, \"Blind separation of sources ~ a nonlinear neural algorithm,\" Neural Nehvorkr, Vol. 5, No, 6, pp. 937-947, 1992.", |
| "links": null |
| }, |
| "BIBREF5": { |
| "ref_id": "b5", |
| "title": "An information-maximisation approach to blind separation and blind deconvolution", |
| "authors": [ |
| { |
| "first": "A", |
| "middle": [ |
| "J" |
| ], |
| "last": "Bell", |
| "suffix": "" |
| }, |
| { |
| "first": "T", |
| "middle": [ |
| "J" |
| ], |
| "last": "Sejnowski", |
| "suffix": "" |
| } |
| ], |
| "year": 1995, |
| "venue": "Neural Computation", |
| "volume": "7", |
| "issue": "6", |
| "pages": "1004--1034", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "A. J. Bell and T. J. Sejnowski, \"An information-maximisation approach to blind separation and blind deconvolution,\" Neural Computation, Vol. 7, No. 6, 1004-1034, 1995.", |
| "links": null |
| }, |
| "BIBREF6": { |
| "ref_id": "b6", |
| "title": "Information theoretic Approaches to source separation", |
| "authors": [ |
| { |
| "first": "P", |
| "middle": [], |
| "last": "Smaragdis", |
| "suffix": "" |
| } |
| ], |
| "year": 1997, |
| "venue": "MIT", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "P . Smaragdis, Information theoretic Approaches to source separation, Master's Thesis, MIT, Cambridge, MA, 1997.", |
| "links": null |
| }, |
| "BIBREF7": { |
| "ref_id": "b7", |
| "title": "Faithful representation of separable distributions", |
| "authors": [ |
| { |
| "first": "I", |
| "middle": [], |
| "last": "Lin", |
| "suffix": "" |
| }, |
| { |
| "first": "D", |
| "middle": [], |
| "last": "Grier", |
| "suffix": "" |
| }, |
| { |
| "first": "J", |
| "middle": [], |
| "last": "Cowan", |
| "suffix": "" |
| } |
| ], |
| "year": 1997, |
| "venue": "Neural Computation", |
| "volume": "9", |
| "issue": "", |
| "pages": "1305--1320", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "I. Lin, D. Grier, and J. Cowan, \"Faithful representation of separable distributions,\" Neural Computation, Vol. 9, pp. 1305-1320,1997.", |
| "links": null |
| }, |
| "BIBREF8": { |
| "ref_id": "b8", |
| "title": "A semi-blind approach to the separation of real world speech mixtures", |
| "authors": [ |
| { |
| "first": "F", |
| "middle": [], |
| "last": "Tordini", |
| "suffix": "" |
| }, |
| { |
| "first": "F", |
| "middle": [], |
| "last": "Piazza", |
| "suffix": "" |
| } |
| ], |
| "year": 2002, |
| "venue": "IJCNN'02", |
| "volume": "2", |
| "issue": "", |
| "pages": "1293--1298", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "F. Tordini and F. Piazza, \"A semi-blind approach to the separation of real world speech mixtures,\" in IJCNN'02, Vol. 2, 2002, pp. 1293-1298.", |
| "links": null |
| }, |
| "BIBREF9": { |
| "ref_id": "b9", |
| "title": "Independent component analysis", |
| "authors": [ |
| { |
| "first": "A", |
| "middle": [], |
| "last": "Hyv\u00e4rinen", |
| "suffix": "" |
| }, |
| { |
| "first": "J", |
| "middle": [], |
| "last": "Karhunen", |
| "suffix": "" |
| }, |
| { |
| "first": "E", |
| "middle": [], |
| "last": "Oja", |
| "suffix": "" |
| } |
| ], |
| "year": 2001, |
| "venue": "", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "A. Hyv\u00e4rinen, J. Karhunen, and E. Oja. Independent component analysis. Wiley, 1st edition, 2001.", |
| "links": null |
| }, |
| "BIBREF10": { |
| "ref_id": "b10", |
| "title": "Statistical Inference", |
| "authors": [ |
| { |
| "first": "Roger", |
| "middle": [ |
| "L" |
| ], |
| "last": "Berger", |
| "suffix": "" |
| }, |
| { |
| "first": "George", |
| "middle": [], |
| "last": "Casella", |
| "suffix": "" |
| } |
| ], |
| "year": 2002, |
| "venue": "", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Roger L.berger, George Casella , Statistical Inference. 2nd edition , DUXBURY 2002.", |
| "links": null |
| }, |
| "BIBREF12": { |
| "ref_id": "b12", |
| "title": "New approcimations of differential entropy for independent component analysis and projection pursuit", |
| "authors": [ |
| { |
| "first": "Aapo", |
| "middle": [], |
| "last": "Hyv\u00e4rinen", |
| "suffix": "" |
| } |
| ], |
| "year": 1998, |
| "venue": "Advance Neural Inform. Processing Syst", |
| "volume": "10", |
| "issue": "", |
| "pages": "273--279", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Aapo Hyv\u00e4rinen, \"New approcimations of differential entropy for independent component analysis and projection pursuit,\" Advance Neural Inform. Processing Syst. 10. MIT Press, pp.273-279, 1998.", |
| "links": null |
| }, |
| "BIBREF13": { |
| "ref_id": "b13", |
| "title": "Partially blind source separation of continuous chaotic signals from linear mixture", |
| "authors": [ |
| { |
| "first": "W", |
| "middle": [], |
| "last": "Hu", |
| "suffix": "" |
| }, |
| { |
| "first": "Z", |
| "middle": [], |
| "last": "Liu", |
| "suffix": "" |
| } |
| ], |
| "year": 2008, |
| "venue": "The Institution of Engineering and Technology", |
| "volume": "2", |
| "issue": "", |
| "pages": "424--430", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "W. Hu, and Z. Liu, \"Partially blind source separation of continuous chaotic signals from linear mixture,\" The Institution of Engineering and Technology 2008, Vol. 2, No4, pp. 424-430.", |
| "links": null |
| }, |
| "BIBREF14": { |
| "ref_id": "b14", |
| "title": "Blind extraction of chaotic signal from an instantaneous linear mixture", |
| "authors": [ |
| { |
| "first": "B", |
| "middle": [ |
| "Y" |
| ], |
| "last": "Wang", |
| "suffix": "" |
| }, |
| { |
| "first": "W", |
| "middle": [ |
| "X" |
| ], |
| "last": "Zheng", |
| "suffix": "" |
| } |
| ], |
| "year": 2006, |
| "venue": "Circuits Syst II", |
| "volume": "53", |
| "issue": "2", |
| "pages": "143--147", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Wang B.Y., and Zheng W.X., \"Blind extraction of chaotic signal from an instantaneous linear mixture,\" Circuits Syst II, 2006, 53, (2), pp.143-147.", |
| "links": null |
| }, |
| "BIBREF15": { |
| "ref_id": "b15", |
| "title": "Separaion and synchronization of piecewise linear chaotic systems", |
| "authors": [ |
| { |
| "first": "A", |
| "middle": [], |
| "last": "Paolo", |
| "suffix": "" |
| }, |
| { |
| "first": "B", |
| "middle": [], |
| "last": "Arturo", |
| "suffix": "" |
| }, |
| { |
| "first": "F", |
| "middle": [], |
| "last": "Luigi", |
| "suffix": "" |
| }, |
| { |
| "first": "F", |
| "middle": [], |
| "last": "Mattia", |
| "suffix": "" |
| } |
| ], |
| "year": 2006, |
| "venue": "Phys. Rev. E", |
| "volume": "74", |
| "issue": "", |
| "pages": "26212--26213", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Paolo A., Arturo B., Luigi F., and Mattia F., \"Separaion and synchronization of piecewise linear chaotic systems,\" Phys. Rev. E, 2006, 74, p. 026212-1-026212-11.", |
| "links": null |
| } |
| }, |
| "ref_entries": { |
| "TABREF0": { |
| "html": null, |
| "text": "\u95dc\u9375\u8a5e\uff1a\u8a9e\u97f3\u8fa8\uf9fc\uff0c\u76f2\u8a0a\u865f\u5206\uf9ea\u6cd5\uff0c\u7368\uf9f7\u6210\u5206\u5206\u6790\uff0c\u5b50\u7a7a\u9593\u8a9e\u97f3\u589e\u5f37\uff0c\u9ea5\u514b\u98a8\u9663\uf99cKeywords: Speech Recognition, Blind Source S eparation, Independent Component Analysis, Subspace Speech Enhancement, Microphone Array.", |
| "content": "<table><tr><td>\u5b78\u6240\u63d0\u4f9b\u7684 HTK(Hidden Markov Model T oolkit)\u8a9e\u97f3\u5957\u4ef6\uf92d\u9032\ufa08\uf9fc\u5225\uff0c\u4e26\u5224\u65b7\u6240\u7522\u751f\u7684</td></tr><tr><td>\u7d50\u679c\u662f\u5426\u6b63\u78ba\u3002</td></tr><tr><td>\u672c\uf941\u6587\u7e3d\u5171\u5206\u6210\u4e94\u500b\u7ae0\u7bc0\uff0c\u7b2c\u4e00\u7ae0\u7bc0\u70ba\u7dd2\uf941\uff0c\u7b2c\u4e8c\u7ae0\u7bc0\u70ba\u672c\uf941\u6587\u91dd\u5c0d\u6b64\u8fa8\uf9fc\u7cfb\u7d71\u6240</td></tr><tr><td>\u63a1\u7528\u4e4b\u5404\u7a2e\u7814\u7a76\u65b9\u6cd5\u4e26\u8a73\u7d30\u52a0\u4ee5\u4ecb\u7d39\uff0c\u7b2c\u4e09\u7ae0\u7bc0\u5247\u662f\u4ecb\u7d39\u6b64\u8a9e\u97f3\u8fa8\uf9fc\u7cfb\u7d71\u4e4b\u7cfb\u7d71\u67b6\u69cb\uff0c</td></tr><tr><td>\u7b2c\u56db\u7ae0\u7bc0\u5247\u662f\u5be6\u9a57\u74b0\u5883\u8a55\u4f30\u548c\u8a2d\u5b9a\u4ee5\u53ca\u5be6\u9a57\u7d50\u679c\uff0c\u6700\u5f8c\u7b2c\u4e94\u7ae0\u7bc0\u5247\u662f\u5c0d\u6b64\u8fa8\uf9fc\u7cfb\u7d71\u505a\u4e00</td></tr><tr><td>\u4e00\u3001\u7dd2\uf941 \u7cbe\u8981\u7d50\uf941\u53ca\u672a\uf92d\u76f8\u95dc\u5de5\u4f5c\u3002</td></tr><tr><td>\u4e8c\u3001\u7814\u7a76\u65b9\u6cd5</td></tr><tr><td>\u8a9e\u8a00\u70ba\u4eba\uf9d0\u5f7c\u6b64\u6e9d\u901a\u6642\uff0c\u6700\u539f\u59cb\u540c\u6a23\u4e5f\u662f\u6700\u6709\u6548\u7684\u65b9\u5f0f\uff0c\u5728\u79d1\u6280\u84ec\u52c3\u767c\u5c55\u7684\u73fe\u4eca\uff0c</td></tr><tr><td>\u5982\u4f55\u4f7f\u96fb\u8166\u8fa8\uf9fc\u4eba\uf9d0\u8a9e\u8a00\u4e5f\u6210\u70ba\u8a9e\u97f3\u8655\uf9e4\u4e0a\u91cd\u8981\u8b70\u984c\u5176\u4e2d\u4e4b\u4e00\uff0c\u56e0\u6b64\u5c0d\u65bc\u8a9e\u97f3\u8fa8\uf9fc\u7cfb \u7d71\uff0c\u5982\u4f55\u9054\u5230\u6709\u6548\u4e14\u7cbe\u78ba\u7684\u8fa8\uf9fc\u7d50\u679c\uff0c\u4e5f\u662f\u76ee\u524d\u8a9e\u97f3\u8655\uf9e4\uf9b4\u57df\u4e2d\u71b1\u9580\u7684\u7814\u7a76\u8b70\u984c\u3002 \u672c\u7ae0\u7bc0\u91dd\u5c0d\u6b64\u9060\u8ddd\uf9ea\u96dc\u8a0a\u8a9e\u97f3\u8fa8\uf9fc\u7cfb\u7d71\u6240\u63a1\u7528\u7684\u5404\u7a2e\u65b9\u6cd5\uf92d\u52a0\u4ee5\u8a73\u8ff0\uf96f\u660e\u4ecb\u7d39\u3002</td></tr><tr><td>\u5c0d\u65bc\u8a9e\u97f3\u8fa8\uf9fc\u7d50\u679c\uff0c\u5f71\u97ff\u8a9e\u97f3\u8fa8\uf9fc\u7d50\u679c\u7684\u76f8\u95dc\u56e0\u7d20\u5f88\u591a\uff0c\u9019\u4e9b\u76f8\u95dc\u56e0\u7d20\u90fd\u6703\u9020\u6210\u8a9e (\u4e00)\u7368\uf9f7\u6210\u5206\u5206\u6790\u6cd5(Independent Component Analysis, ICA)</td></tr><tr><td>\u8005\u8a9e\u610f\u548c\u8a9e\u97f3\u8fa8\uf9fc\u7d50\u679c\u7684\uf967\u5339\u914d(mismatch)\uff0c\u5176\u4e2d\u5f71\u97ff\u8fa8\uf9fc\u7d50\u679c\u6700\u91cd\u8981\u7684\u56e0\u7d20\u70ba\u74b0\u5883\u4e2d \u5c0d\u65bc\u5e36\u6709\u566a\u8072\u7684\u8a9e\u97f3\u6210\u5206\uff0c\u7531\u65bc\u539f\u59cb\u8a9e\u97f3\u6210\u5206\u548c\u80cc\u666f\u96dc\u8a0a\u6210\u5206\u5747\u70ba\u672a\u77e5\uff0c\u56e0\u6b64\u8981\u5206\uf9ea\u6b64 \u6240\u5b58\u5728\u7684\u80cc\u666f\u96dc\u8a0a\uff0c\u7531\u65bc\u8a9e\u97f3\u6240\u5b58\u5728\u7684\u80cc\u666f\u74b0\u5883\u4e2d\uff0c\u4e26\u975e\u5b8c\u5168\u6c92\u6709\u906d\u53d7\u5176\u4ed6\u5e72\u64fe\u96dc\u8a0a\u5f71 \uf978\u7a2e\u672a\u77e5\u8a0a\u865f\uff0c\u6211\u5011\u53ef\u4f7f\u7528\u76f2\u8a0a\u865f\u5206\uf9ea\u65b9\u5f0f\uff0c\u5c07\u6b64\uf978\u7a2e\u672a\u77e5\u8a0a\u865f\uff0c\u5206\u5225\u5f9e\u6df7\u5408\u8a0a\u865f\u4e2d\u5206 \u97ff\uff0c\uf9b5\u5982\u5728\u9910\u5ef3\u74b0\u5883\u3001\u5730\u9435\u74b0\u5883\u3001\uf902\u5167\ufa08\u99db\u74b0\u5883\u7b49\uff0c\u90fd\u6709\u80cc\u666f\u96dc\u8a0a\u7684\u5e72\u64fe\u6e90\u5b58\u5728\uff0c\u9019\u4e9b \u80cc\u666f\u96dc\u8a0a\u4f34\u96a8\u8457\u8a9e\u97f3\u9032\u5165\u8fa8\uf9fc\u7cfb\u7d71\u4e2d\uff0c\u6703\u56b4\u91cd\u5f71\u97ff\u5230\u6574\u9ad4\u8fa8\uf9fc\u7d50\u679c\uff0c\u53e6\u5916\u8a9e\u8005\u8207\u8fa8\uf9fc\u7cfb \uf9ea\u51fa\uf92d\uff0c\u4e00\u822c\u76f2\u8a0a\u865f\u5206\uf9ea\u554f\u984c\u53ef\u7531\u4e0b\u9762\u793a\u610f\u5716\u8868\u793a\uff1a</td></tr><tr><td>\u7d71\u8ddd\uf9ea\u4e5f\u662f\u53e6\u4e00\u7a2e\u5f71\u97ff\u8fa8\uf9fc\u7d50\u679c\u7684\u56e0\u7d20\uff0c\u8a9e\u97f3\u80fd\uf97e\u6703\u4f34\u96a8\u8457\u8ddd\uf9ea\u800c\u9010\u6f38\u8870\u6e1b\uff0c\u56e0\u6b64\u8870\u6e1b</td></tr><tr><td>\u5f8c\u7684\u8a9e\u97f3\u80fd\uf97e\u4e5f\u6703\u9020\u6210\u8fa8\uf9fc\uf961\u7684\ufa09\u4f4e\u3002</td></tr><tr><td>\u5716\u4e8c\u3001\u76f2\u8a0a\u865f\u5206\uf9ea\u554f\u984c\u793a\u610f\u5716</td></tr><tr><td>\u5982\u5716\u4e8c\u6240\u793a\uff0c\uf978\u672a\u77e5\u8072\u6e90\u8a0a\u865f s 1 \u53ca s 2 \uff0c\u900f\u904e\u6df7\u5408\u77e9\u9663 A \u5f8c\uff0c\u5728\u9ea5\u514b\u98a8\u63a5\u6536\u7aef\u5247\u6703\u63a5\u6536</td></tr><tr><td>\u5230\uf978\u7a2e\u6df7\u5408\u8a0a\u865f x 1 \u548c x 2 \uff0c\u6b64\u95dc\u4fc2\u53ef\u7531\u4e0b\uf99c\u7dda\u6027\u65b9\u7a0b\u5f0f\u8868\u793a\u3002</td></tr><tr><td>\u5716\u4e00\u3001\u80cc\u666f\u74b0\u5883\u96dc\u8a0a\u5e72\u64fe\u8a9e\u97f3\u793a\u610f\u5716 \u70ba\uf9ba\u6539\u5584\u4e0a\u8ff0\u6240\u63d0\u5230\u4e4b\u74b0\u5883\u96dc\u8a0a\u4ee5\u53ca\u8a9e\u8005\u8ddd\uf9ea\u6240\u9020\u6210\u7684\u8fa8\uf9fc\u7d50\u679c\uf967\u5339\u914d\uff0c\u6211\u5011\u91dd\u5c0d 2 12 1 11 1 s a s a x \uf02b \uf03d</td></tr><tr><td>\u6b64\u96dc\u8a0a\u8a9e\u97f3\u505a\u9032\u4e00\u6b65\u5206\u6790\uff0c\u9996\u5148\u96dc\u8a0a\u8a9e\u97f3\u4e2d\u5305\u542b\uf9ba\u5927\uf97e\u7684\u96dc\u8a0a\u8cc7\u8a0a\uff0c\u56e0\u6b64\u5982\u4f55\u53d6\u5f97\u96dc\u8a0a \u90e8\u4efd\u4e26\u52a0\u4ee5\u53bb\u9664\u70ba\u7b2c\u4e00\u6b65\u91cd\u8981\u7684\u8655\uf9e4\u6b65\u9a5f\uff0c\uf984\u9664\u76f8\u95dc\u7684\u80cc\u666f\u96dc\u8a0a\u5f8c\uff0c\u518d\uf92d\u5247\u662f\u8a9e\u8005\u548c\u8fa8 2 22 1 21 2 s a s a x \uf02b \uf03d ,</td></tr><tr><td>\uf9fc\u7cfb\u7d71\u4e4b\u9593\u7684\u8ddd\uf9ea\u554f\u984c\uff0c\u7576\u8ddd\uf9ea\u76f8\u8ddd\u8d8a\u5927\u6642\uff0c\u8a9e\u97f3\u8fa8\uf9fc\u7cfb\u7d71\u6240\u63a5\u6536\u5230\u7684\u8a9e\u97f3\u80fd\uf97e\u5247\u8d8a</td></tr><tr><td>\u5c0f\uff0c\u56e0\u6b64\u5c0d\u65bc\uf984\u9664\u96dc\u8a0a\u5f8c\u7684\u8a9e\u97f3\u8a0a\u865f\uff0c\u5fc5\u9808\u518d\u9032\u4e00\u6b65\u4f7f\u7528\u8a9e\u97f3\u589e\u5f37\u6280\u8853\u5c07\u52a0\u5f37\u8a9e\u97f3\u8a0a\u865f</td></tr><tr><td>\u80fd\uf97e\uff0c\u4ee5\u63d0\u5347\u4e4b\u5f8c\u7684\u8fa8\uf9fc\u7d50\u679c\uff0c\u6700\u5f8c\u5728\u9032\ufa08\u8fa8\uf9fc\u4e4b\u524d\uff0c\u518d\u5c07\u589e\u5f37\u5f8c\u4e4b\u8a9e\u97f3\u8a0a\u865f\u505a\u7aef\u9ede\u5075</td></tr><tr><td>\u6e2c\u8655\uf9e4\uff0c\u627e\u51fa\u4e00\u6bb5\u8a9e\u97f3\u8a0a\u865f\u4e2d\u8a9e\u97f3\u7684\u5be6\u969b\u4f4d\u7f6e\u518d\u53d6\u5f97\u6b64\u8a9e\u97f3\u8cc7\u8a0a\uf92d\u9032\ufa08\u8fa8\uf9fc\u3002</td></tr><tr><td>\u6839\u64da\u4e0a\u8ff0\u5206\u6790\u7d50\u679c\uff0c\u5728\u96dc\u8a0a\u5206\uf9ea\u90e8\u4efd\uff0c\u6211\u5011\u63a1\u7528\u76f2\u8a0a\u865f\u5206\uf9ea(Blind Signal Separation,</td></tr><tr><td>BSS)\u7684\u65b9\u6cd5\uff0c\u4f7f\u7528\u7368\uf9f7\u6210\u5206\u5206\u6790(Independent Component Analysis, ICA)\u65b9\u5f0f\uf92d\u9032\ufa08\u8a0a\u865f</td></tr><tr><td>\u5206\uf9ea\uff0c\u53d6\u51fa\u76f8\u8fd1\u4f3c\u8a9e\u97f3\u6210\u5206\u8f03\u591a\u7684\u90e8\u4efd\uff0c\u518d\u900f\u904e\u5b50\u7a7a\u9593\u8a9e\u97f3\u589e\u5f37\u65b9\u5f0f(Subspace Speech</td></tr><tr><td>Enhancement)\uff0c\u5c07\u53d6\u51fa\u7684\u8a9e\u97f3\u8a0a\u865f\u9032\u4e00\u6b65\u53bb\u9664\u6b98\u9918\u566a\u8072\u4e26\u52a0\u5f37\u8a9e\u97f3\u8a0a\u865f\uff0c\u4f7f\u5176\u53ef\u7528\uf92d\u9032</td></tr><tr><td>\ufa08\u8a9e\u97f3\u8fa8\uf9fc\u4e4b\u7528\uff0c\u6700\u5f8c\u518d\uf9dd\u7528\u8a9e\u97f3\u6d3b\u52d5\u5075\u6e2c\u6cd5(Voice Activity Detection, VAD)\uf92d\u5075\u6e2c\u8a9e\u97f3</td></tr><tr><td>\u6240\u5728\u4f4d\u7f6e\uff0c\u85c9\u6b64\uf92d\u63d0\u5347\u8fa8\uf9fc\u6548\uf961\u3002\u6700\u5f8c\u5728\u672b\u7aef\u7684\u8a9e\u97f3\u8fa8\uf9fc\u5668\u65b9\u9762\uff0c\u6211\u5011\u4f7f\u7528\u82f1\u570b\u528d\u6a4b\u5927</td></tr></table>", |
| "num": null, |
| "type_str": "table" |
| }, |
| "TABREF1": { |
| "html": null, |
| "text": "\u516c\u5c3a\u4e14\u9ad8\ufa01\u70ba 75 \u516c\u5206\uff0c\u4e14\uf978\u652f\u6536\u97f3\u9ea5\u514b\u98a8\u9593\u8ddd\u70ba 14 \u516c\u5206\uff0c\u9ad8\ufa01\u70ba 55 \u516c\u5206\u3002\u5728\u8a9e\u8005\u90e8\u4efd\uff0c\u6211\u5011\u63a1\u7528\u4e09\u4eba\u9032\ufa08\uf93f\u97f3\uff0c\u4e14\u6bcf\u4eba\u5404\uf96f 10 \uf906\u4e09\u5b57\u8a5e\u9032\ufa08\uf93f\u88fd\uff0c\u5728\u6b64 \u6211\u5011\u4ee5\u4eba\u540d\u505a\u70ba\u5b57\u8a5e\uf92d\u6e90\uff1b\u5728\u566a\u97f3\u90e8\u4efd\uff0c\u6211\u5011\u63a1\u7528 noise-92 \u6240\u63d0\u4f9b\u7684\u566a\u8072\u8cc7\uf9be\u5eab\u4f5c\u70ba\u566a \u8072\uf92d\u6e90\uff0c\u5728\u5be6\u9a57\u4e2d\u6211\u5011\u4f7f\u7528\uf967\u540c\u566a\u97f3\u6bb5\u7684 babble noise \u548c car noise \u7576\u4f5c\u566a\u8072\u7a2e\uf9d0\u3002 \u8a9e\u97f3\u548c\u566a\u97f3\u6df7\u5408\u5f8c\u7684\u8a0a\u865f\u90e8\u4efd\uff0c\u6211\u5011\u6839\u64da SNR(Signal-to-noise ratio)\uff0c\u5206\u5225\u7522\u751f\u5404\u7a2e\u566a\u8072 \u60c5\u5883\u4e0b\u4e09\u7a2e\uf967\u540c SNR \u503c\u7684\u96dc\u8a0a\u8a0a\u865f\uff0c\u5206\u5225\u662f 0dB\u30015dB\u3001\u4ee5\u53ca 10dB\uff0cSNR \u516c\u5f0f\u5982(24) \u5f0f\u6240\u793a\uff0c\u5176\u4e2d P signal \u548c P noise \u5206\u5225\u6307\u8a0a\u865f\u548c\u96dc\u8a0a\u7684\u5e73\u5747\u80fd\uf97e\uff0cA signal \u548c A noise \u5247\u662f\u6307\u8a0a\u865f\u548c \u96dc\u8a0a\u632f\u798f\u5927\u5c0f\uff0c\u6700\u5f8c\u518d\u5c07\u5404\u7a2e\u60c5\u5883\u4e0b\u7684\u6df7\u5408\u96dc\u8a0a\u8a0a\u865f\u9032\ufa08\u8a9e\u97f3\u5206\uf9ea\u548c\u8a9e\u97f3\u589e\u5f37\uff0c\u6700\u5f8c\u518d", |
| "content": "<table><tr><td colspan=\"6\">18) \u85c9\u7531(17)\u5f0f\u548c(18)\u5f0f\u518d\u548c(16)\u5f0f\u76f8\u6bd4\u8f03\uff0c\uf974\u8981\u5c0d\u8a0a\u865f\u5b50\u7a7a\u9593\u4e2d\u7684\uf984\u6ce2\u5668\u4f5c\u6700\u4f73\u5316\u8655\uf9e4\uff0c\u5c0d \u65bc\u8a9e\u97f3\u8a0a\u865f\u90e8\u4efd\uff0c\u8a9e\u97f3\u5931\u771f\u7684\u7a0b\ufa01\u8981\u6700\u5c0f\uff0c\u5c0d\u65bc\u566a\u97f3\u8a0a\u865f\u90e8\u4efd\uff0c\u6b98\uf9cd\u7684\u566a\u97f3\u53ea\u8981\u76e1\uf97e\u6291 \u5236\u5230\uf967\u81f3\u65bc\u5f71\u97ff\u8fa8\uf9fc\u7d50\u679c\u7684\u7a0b\ufa01\u5c31\u597d\uff0c\u800c\u975e\u8981\u6c42\u5b8c\u5168\u6c92\u6709\u6b98\u5b58\u7684\u566a\u97f3\u6210\u5206\u5b58\u5728\uff0c\u5728\u5982\u6b64 \u6298\u8877\u7684\u689d\u4ef6\u4e0b\u6211\u5011\u5c07\u6b64\uf984\u6ce2\u5668\u7684\u6700\u4f73\u5316\u689d\u4ef6\u4ee5(19)\u5f0f\uf92d\u8868\u793a\u3002 1 0 , min 2 \uf0a3 \uf0a3 \uf0a3 \uf067 \uf067\uf073 \uf064 \uf064 S n y (19) \u5176\u4e2d\u03c3 2 \u70ba\u566a\u97f3\u7684\u8b8a\uf962\uf969\uff0c\u03b3\u70ba\u8abf\u6574\u63a7\u5236\uf984\u6ce2\u5668\u6b98\uf9cd\u566a\u97f3\u8a0a\u865f\u7684\u7a0b\ufa01\uff0c\u56e0\u6b64\u6211\u5011\u4f7f\u7528 Lagrange \u65b9\u6cd5\uf92d\u8a08\u7b97\u6b64\u6700\u4f73\u5316\uf984\u6ce2\u5668\u689d\u4ef6\uff0c\u63a8\u5c0e\u7d50\u679c\u5982\u4e0b\uff0c\u03bc\u70ba Lagrange m ultiplier\uff0c R y \u548c \u5206\u5225\u70ba\u8a9e\u97f3\u8a0a\u865f\u548c\u566a\u97f3\u8a0a\u865f\u7684\u5171\u8b8a\uf962\u77e9\u9663\uff0c\uf974\u5c07 R y \u4f7f\u7528\u7279\u5fb5\u503c\u5206\u89e3\uff0c\u5047\u8a2d R 21)\u5f0f\uff0c\u6700\u5f8c \u518d\u5c07 \u7528\u566a\u97f3\u8a0a\u865f\u7684\u7279\u5fb5\u503c\u5c0d\u89d2\u77e9\u9663\u8fd1\u4f3c\uff0c\u5f97\u5230\u6700\u5f8c\u8a0a\u865f\u5b50\u7a7a\u9593\uf984\u6ce2\u5668\u6700\u4f73\u5316\u8655 \uf9e4\u7d50\u679c\u3002 S n R R P n T P S (20) 1 ) ( \uf02d \uf02b \uf03d S n y y R R R F \uf06d (21) T n T y y P P R P D PD F S 1 ) ( \uf02d \uf02b \uf03d \uf06d (22) T n y y P D D PD F S 1 ) ( \uf02d \uf02b \uf03d \uf06d (\u4e09)\u8a9e\u97f3\u6d3b\u52d5\u5075\u6e2c\u6cd5(Voice Activity Detection, VAD) \u5728\u8a9e\u97f3\u6d3b\u52d5\u5075\u6e2c\u6cd5\u4e0a\uff0c\u6211\u5011\uf9dd\u7528\u8a9e\u97f3\u8a0a\u865f\u7684\u80fd\uf97e\u66f2\u7dda\u548c\u904e\uf9b2\uf961(Zero crossing rate)\uf92d\u9032\ufa08 \u8a9e\u97f3\u8a0a\u865f\u7684\u7aef\u9ede\u5075\u6e2c\uff0c\u4e00\u958b\u59cb\u6211\u5011\u9810\u5148\u5728\u8a9e\u97f3\u8a0a\u865f\u6ce2\u5f62\u4e0a\u8a2d\u5b9a\u4e00\u689d\u57fa\u6e96\u7dda\uff0c\u7576\u8a0a\u865f\u632f\u5e45 \u5728\u6b64\u57fa\u6e96\u7dda\u4e0a\u65b9\u5b9a\u7fa9\u70ba\u6b63\uff0c\u53cd\u4e4b\u5b9a\u7fa9\u70ba\u8ca0\uff0c\u518d\uf92d\u5247\u91dd\u5c0d\u8a0a\u865f\u4e2d\u6bcf\u500b\u97f3\u6846\uff0c\u500b\u5225\u8a08\u7b97\u632f\u5e45 \u7531\u6b63\u5230\u8ca0\u3001\u4ee5\u53ca\u7531\u8ca0\u5230\u6b63\u7684\u6b21\uf969\uff0c\uf974\u55ae\u4f4d\u6642\u9593\u5167\u8d8a\u904e\u57fa\u6e96\u7dda\u6b21\uf969\u589e\u591a\uff0c\u8868\u793a\u8a0a\u865f\u6ce2\u5f62\u64fa \u52d5\u8d8a\u5287\uf99f\u3002\u5c0d\u65bc\u4e00\u6bb5\u542b\u96dc\u8a0a\u4e4b\u8a9e\u97f3\u8a0a\u865f\uff0c\u96dc\u8a0a\u6216\u6c23\u9f3b\u97f3\u80fd\uf97e\u8f03\u5c0f\u4e14\u904e\uf9b2\uf961\u8f03\u9ad8\uff0c\u800c\u8a9e\u97f3 \u90e8\u4efd\u5247\u662f\u8a9e\u97f3\u80fd\uf97e\u8f03\u9ad8\u4e14\u904e\uf9b2\uf961\u4f4e\uff0c\u56e0\u6b64\u53ef\u85c9\u7531\u80fd\uf97e\u66f2\u7dda\u4ee5\u53ca\u904e\uf9b2\uf961\uf92d\u91dd\u5c0d\u6bcf\u6bb5\u8a9e\u97f3\u8a0a \u5176\u4e2d x(n)\u8868\u793a\u7b2c n \u500b\u6a23\u672c\u9ede\u7684\u632f\u5e45\u80fd\uf97e\uff0cx(n-1)\u8868\u793a\u70ba\u524d\u4e00\u500b\u6a23\u672c\u9ede\uff0c\u56e0\u6b64\u904e\uf9b2\uf961\u662f\u6307 \uf978\uf99a\u7e8c\u6a23\u672c\u9593\uff0c\u5177\u6709\uf967\u540c\u7684\u6b63\u8ca0\u865f\u6b21\uf969\u3002\u53d6\u51fa\u6b63\u78ba\u7684\u8a9e\u97f3\u8a0a\u865f\u5f8c\u5373\u53ef\u958b\u59cb\u9032\ufa08\u8fa8\uf9fc\uff0c\u4e0b \u5716\u70ba\u8a9e\u97f3\u6d3b\u52d5\u5075\u6e2c\u6cd5\u6574\u9ad4\uf9ca\u7a0b\u5716\u3002 \u5716\u4e09\u3001\u8a9e\u97f3\u6d3b\u52d5\u5075\u6e2c\u6cd5\uf9ca\u7a0b\u5716 \u4e09\u3001\u7cfb\u7d71\u67b6\u69cb \u5728\u4e0a\u4e00\u7ae0\u7bc0\u4e2d\u6211\u5011\u8a73\u7d30\u6558\u8ff0\u672c\uf941\u6587\u6240\u63d0\u51fa\u7684\u9060\u8ddd\uf9ea\u96dc\u8a0a\u8a9e\u97f3\u8fa8\uf9fc\u7cfb\u7d71\uff0c\u6240\u63a1\u7528\u7684\u5404\u7a2e\u7814 \u7a76\u65b9\u6cd5\uff0c\u7576\u6536\u97f3\u7cfb\u7d71\u63a5\u6536\u5230\u542b\u6709\u80cc\u666f\u96dc\u8a0a\u7684\u8a9e\u97f3\u6642\uff0c\u9996\u5148\u6703\u7d93\u7531\u76f2\u8a0a\u865f\u5206\uf9ea\u6240\u4f7f\u7528\u7684\u7368 \uf9f7\u6210\u5206\u5206\u6790\u6cd5\u5c07\u5e36\u6709\u96dc\u8a0a\u7684\u6df7\u5408\u8a0a\u865f\u9032\ufa08\u5206\uf9ea\uff0c\u5206\uf9ea\u51fa\uf978\u500b\u7368\uf9f7\u8a0a\u865f\uff0c\u518d\u5f9e\u9019\uf978\u500b\u7368\uf9f7 \u8a0a\u865f\u4e2d\u9078\u53d6\u8a9e\u97f3\u6210\u5206\u8f03\u591a\u7684\u7368\uf9f7\u8a0a\u865f\uff0c\u4f7f\u7528\u5b50\u7a7a\u9593\u8a9e\u97f3\u589e\u5f37\u6cd5\u9032\u4e00\u6b65\uf984\u9664\u8a0a\u865f\u4e2d\u96dc\u8a0a\u6210 \u5206\uff0c\u6700\u5f8c\u518d\uf9dd\u7528\u8a9e\u97f3\u6d3b\u52d5\u5075\u6e2c\u6cd5\u9032\ufa08\u7aef\u9ede\u5075\u6e2c\uff0c\u6700\u5f8c\u518d\u4f7f\u7528 HTK \u8a9e\u97f3\u5957\u4ef6\u9032\ufa08\u8fa8\uf9fc\uff0c \u4e26\u5224\u65b7\u5176\u8fa8\uf9fc\u7d50\u679c\u662f\u5426\u6b63\u78ba\uff0c\u4e0b\u5716\u70ba\u6574\u9ad4\u9060\u8ddd\uf9ea\u96dc\u8a0a\u8a9e\u97f3\u8fa8\uf9fc\u7cfb\u7d71\u6574\u9ad4\u67b6\u69cb\uf9ca\u7a0b\u5716\u4ee5\u53ca HTK \u8a9e\u97f3\u8fa8\uf9fc\u5668\u8fa8\uf9fc\uf9ca\u7a0b\u5716\u3002 \u5716\u56db\u3001\u9060\u8ddd\uf9ea\u96dc\u8a0a\u8a9e\u97f3\u8fa8\uf9fc\u7cfb\u7d71\uf9ca\u7a0b\u5716 \u56db\u3001\u5be6\u9a57\u8a2d\u5b9a\u53ca\u8fa8\uf9fc\u7d50\u679c (\u4e00)\u5be6\u9a57\u74b0\u5883\u8a55\u4f30\u548c\u60c5\u5883\u8a2d\u5b9a \u5728\u74b0\u5883\u8a55\u4f30\u65b9\u9762\uff0c\u5be6\u9a57\u74b0\u5883\u5982\u4e0b\u5716\u6240\u793a\uff0c\u6703\u8b70\u5ba4\u9ad8\ufa01\u7d04\u70ba 3 \u516c\u5c3a\uff0c\u81f3\u65bc\u6536\u97f3\u9ea5\u514b\u98a8\u9663\uf99c\uff0c \u6211\u5011\u63a1\u7528\uf978\u652f\u9ea5\u514b\u98a8\u9032\ufa08\u6536\u97f3\uff0c\u518d\u6839\u64da\uf967\u540c\u566a\u8072\u74b0\u5883\u53ca\u8a9e\u8005\u8eab\u5206\u548c\u8a9e\u610f\u5167\u5bb9\u9032\ufa08\u8fa8\uf9fc\u3002 \u5716\uf9d1\u3001\u5be6\u9a57\u74b0\u5883\u793a\u610f\u5716 \u6839\u64da\u4e0a\u5716\u5be6\u9a57\u74b0\u5883\u793a\u610f\u5716\uff0c\u6211\u5011\u8a2d\u5b9a\u8a9e\u8005\u8ddd\uf9ea\u9ea5\u514b\u98a8\u9663\uf99c\u4e2d\u5fc3\u70ba 1.5 \u516c\u5c3a\uff0c\u566a\u8072\u6e90\u8ddd\uf9ea ) A A ( 20log ) P P ( 10log SNR(dB) noise signal 10 noise signal 10 \uf03d \uf03d (24) (\u4e8c)\u5be6\u9a57\u6a21\u64ec\u8fa8\uf9fc\u7d50\u679c \u6700\u5f8c\u5728\u5be6\u9a57\u8fa8\uf9fc\u7d50\u679c\u65b9\u9762\uff0c\u6211\u5011\u4ee5\u8a9e\u97f3\u5206\uf9ea\u548c\u589e\u5f37\u5f8c\u7684\u8a0a\u865f\u5e73\u5747 SNR \u503c\uff0c\u4ee5\u53ca\u8207\u539f\u59cb \u4e7e\u6de8\u8a9e\u97f3\u6bd4\u8f03\u7684 Segment SNR \u503c\uff0c\u9084\u6709\u8fa8\uf9fc\uf961(Recognition rate)\u7576\u4f5c\u6211\u5011\u8fa8\uf9fc\u7d50\u679c\u7684\u4e3b \u8981\u4f9d\u64da\uff0c\u5176\u4e2d Segment SNR \u516c\u5f0f\u5982(25)\u5f0f\u6240\u8868\u793a\uff0c\u5176\u4e2d d(i)\u548c y(i)\u5206\u5225\u70ba\u539f\u59cb\u4e7e\u6de8\u8a9e\u97f3 \u8a0a\u865f\u548c\u589e\u76ca\u5f8c\u8a9e\u97f3\u8a0a\u865f\u3002 \uf0e5 \uf0e5 \uf03d \uf03d \uf03d 1 -T 0 t 1 -N 0 i 2 2 10 ] y(i)) -(d(i) (i) d [10log T 1 SegSNR(dB) (25) \u5728\u4e0b\uf99c\u7684\u5be6\u9a57\u8868\u683c\u4e2d\uff0c\u8868\u4e00\u70ba\u4e09\u7a2e\uf967\u540c\u566a\u97f3\u6bb5\u7684 babble noise \u548c car noise \u8207\u8a9e\u97f3\u6240\u6df7\u5408 \u800c\u6210\u7684\u96dc\u8a0a\u8a9e\u97f3\uff0c\u518d\u4f9d\u64da\u4e09\u7a2e\uf967\u540c SNR \u503c(0dB\u30015dB\u300110dB)\u60c5\u6cc1\u4e0b\u9032\ufa08\u6df7\u5408\uff0c\u6700\u5f8c\u518d \u5c07\u6b64\u6df7\u5408\u5f8c\u7684\u96dc\u8a0a\u8a9e\u97f3\u9032\ufa08\u566a\u97f3\u5206\uf9ea\u53ca\u8a9e\u97f3\u589e\u5f37\uff0c\u4e26\u8a08\u7b97\u5176\u589e\u76ca\u5f8c\u8a0a\u865f\u7684\u5e73\u5747 SNR \u503c \u548c Segment SNR \u503c\u3002\u5728\u8868\u4e00\u4e2d\u6211\u5011\u53ef\u6e05\u695a\u770b\ufa0a\uff0c\u589e\u76ca\u5f8c\u7684\u8a9e\u97f3\u8a0a\u865f\uff0c\u7121\uf941\u662f\u5728\u5e73\u5747 SNR \u503c\u6216\u662f Segment SNR \u503c\uff0c\u5747\u6bd4\u539f\u59cb\u5e73\u5747\u503c\u63d0\u5347\uf967\u5c11\uff0c\u6574\u9ad4\u63d0\u5347\u5747\u8d85\u904e 20dB\u3002 \u8868\u4e00\u3001\u5404\u7a2e\u566a\u97f3\u60c5\u5883\u4e0b\u589e\u76ca\u5f8c\u8a0a\u865f SNR \u503c\u8207 SegSNR \u503c \u566a\u97f3\u60c5\u5883 \u539f\u59cb SNR \u503c \u589e\u76ca\u5f8c SNR \u503c \u589e\u76ca\u5f8c SegSNR \u503c Babble noise 1 (0dB\u30015dB\u300110dB) 5 dB 17.99 dB 30.27 dB Babble noise 2 (0dB\u30015dB\u300110dB) 5 dB 21.81 dB 30.08 dB Babble noise 3 (0dB\u30015dB\u300110dB) 5 dB 22.39 dB 31.21 dB Car noise 1 (0dB\u30015dB\u300110dB) 5 dB 28.25 dB 31.79 dB Car noise 2 (0dB\u30015dB\u300110dB) 5 dB 30.76 dB 32.50 dB Car noise 3 (0dB\u30015dB\u300110dB) 5 dB 31.31 dB 33.17 dB \u8868\u4e8c\u5247\u662f\u5728\u7121\u80cc\u666f\u566a\u97f3\u53ca\u5404\u7a2e\u566a\u97f3\u60c5\u5883\u4e0b\uff0c\u539f\u59cb\u96dc\u8a0a\u8a9e\u97f3\u8a0a\u865f\u7684\u8fa8\uf9fc\uf961\u8207\u589e\u76ca\u5f8c\u7684\u8fa8\uf9fc \uf961\u6bd4\u8f03\u8868\uff0c\u6211\u5011\u5728\u8868\u4e8c\u4e2d\u53ef\u770b\ufa0a\uff0c\u589e\u76ca\u904e\u5f8c\u7684\u8a9e\u97f3\u8a0a\u865f\u5728\u8fa8\uf9fc\uf961\u4e0a\u6709\u4e00\u5b9a\u7684\u63d0\u5347\u7a0b\ufa01\uff0c \u8207\u539f\u59cb\u96dc\u8a0a\u8a9e\u97f3\u8fa8\uf9fc\uf961\u76f8\u6bd4\u8f03\uff0c\u6700\u9ad8\u53ef\u63d0\u5347 30%\u8fa8\uf9fc\uf961\uff0c\u6574\u9ad4\u800c\u8a00\u7d04\u53ef\u63d0\u6607 22.96%\u8fa8 \uf9fc\uf961\u3002 \u8868\u4e8c\u3001\u5404\u7a2e\u566a\u97f3\u60c5\u5883\u4e0b\u589e\u76ca\u5f8c\u8a0a\u865f\u8fa8\uf9fc\uf961 \u566a\u97f3\u60c5\u5883 \u539f\u59cb\u8fa8\uf9fc\uf961 \u589e\u76ca\u5f8c\u8fa8\uf9fc\uf961 \u4e0b\u5716\u70ba\u539f\u59cb\u96dc\u8a0a\u8a9e\u97f3\u8a0a\u865f\u7684\u6ce2\u5f62\u5716\u548c\u8a0a\u865f\u8072\u8b5c\u5716(Spectrogram)\u3001\u4ee5\u53ca\u589e\u76ca\u5f8c\u8a9e\u97f3\u8a0a\u865f\u7684 \u6ce2\u5f62\u5716\u548c\u8072\u8b5c\u5716\uff0c\u5728\uf978\u8005\u8a0a\u865f\u7684\u6ce2\u5f62\u5716\u6bd4\u8f03\u4e2d\uff0c\u53ef\u660e\u986f\u770b\ufa0a\u589e\u76ca\u5f8c\u7684\u8a9e\u97f3\u8a0a\u865f\uff0c\u5728\u96dc\u8a0a \u6291\u5236\u4e0a\u6709\u986f\u8457\u7684\u63d0\u5347\uff1b\u5728\u8072\u8b5c\u5716\u6bd4\u8f03\u4e2d\uff0c\u9664\uf9ba\u53ef\u767c\u73fe\u5230\u975e\u8a9e\u97f3\u6bb5\u8a0a\u865f\u80fd\uf97e\u5206\u5e03\u5df2\ufa09\u4f4e\uf967 \u5c11\uff0c\u4e26\u4e14\u8a9e\u97f3\u90e8\u4efd\u7684\u8a0a\u865f\u80fd\uf97e\u4ea6\u63d0\u6607\u8a31\u591a\uff0c\u8b49\u5be6\u672c\uf941\u6587\u6240\u63d0\u51fa\uf92d\u7684\u9060\u8ddd\uf9ea\u96dc\u8a0a\u8a9e\u97f3\u8fa8\uf9fc \u7cfb\u7d71\uff0c\u5177\u6709\uf97c\u597d\uf984\u9664\u566a\u8072\u96dc\u8a0a\u548c\u52a0\u5f37\u8a9e\u97f3\u6210\u5206\u7b49\u529f\u80fd\uff0c\u4e14\u80fd\u6709\u6548\u7684\u63d0\u5347\u5176\u8fa8\uf9fc\uf961\u3002 \u5716\u4e03\u3001\u539f\u59cb\u96dc\u8a0a\u8a9e\u97f3\u8a0a\u865f\u548c\u589e\u76ca\u5f8c\u8a9e\u97f3\u8a0a\u865f\u4e4b\u6ce2\u5f62\u5716\u548c\u8072\u8b5c\u5716 \u4e94\u3001\u7d50\uf941 \u672c\uf941\u6587\u6240\u63d0\u51fa\uf92d\u7684\u9060\u8ddd\uf9ea\u96dc\u8a0a\u8a9e\u97f3\u8fa8\uf9fc\u7cfb\u7d71\uff0c\u4e3b\u8981\u85c9\u7531\u76f2\u8a0a\u865f\u5206\uf9ea\u65b9\u5f0f\u4ee5\u53ca\u8a9e\u97f3\u589e\u5f37\u6280 \u8853\u5c07\u96dc\u8a0a\u8a9e\u97f3\u5206\uf9ea\u51fa\u55ae\u4e00\u7368\uf9f7\u8a9e\u97f3\u8a0a\u865f\uff0c\u518d\u900f\u904e\u8a9e\u97f3\u589e\u5f37\u9032\u4e00\u6b65\uf984\u9664\u8a9e\u97f3\u8a0a\u865f\u4e2d\u6b98\uf9cd\u566a \u97f3\uf92d\u63d0\u5347\u8fa8\uf9fc\uf961\uff0c\u5728\u5be6\u9a57\u7d50\u679c\u986f\u793a\uff0c\u672c\uf941\u6587\u6240\u63d0\u51fa\uf92d\u4e4b\u8fa8\uf9fc\u7cfb\u7d71\uff0c\u53ef\u660e\u986f\u7684\u6709\u6548\u63d0\u5347\u8a9e \u97f3\u80fd\uf97e\u4ee5\u53ca\u8fa8\uf9fc\uf961\uff0c\u672a\uf92d\u6211\u5011\u5c07\u6a21\u64ec\uf901\u591a\uf967\u540c\u4eba\u8072\u53ca\u566a\u97f3\u60c5\u5883\u3001\u63a2\u8a0e\uf901\u591a\uf967\u540c\u7814\u7a76\u65b9 \u9ea5\u514b\u98a8\u4e2d\u5fc3\u70ba 2 \u9032\ufa08\u8fa8\uf9fc\uf9ca\u7a0b\u3002 \u7121\u566a\u97f3 58.89 % 72.22 % \u6cd5\uff0c\u767c\u5c55\u51fa\u4e00\u5957\uf901\u5177\u9ad8\u97f3\u8cea\u89e3\u6790\u4e14\u9ad8\u8fa8\uf9fc\uf961\u4e4b\u9060\u8ddd\uf9ea\u8a9e\u97f3\u8fa8\uf9fc\u7cfb\u7d71\u3002</td></tr><tr><td colspan=\"6\">\u865f\u9032\ufa08\u7aef\u9ede\u5075\u6e2c\u8655\uf9e4\u3002\u5728\u6b64\u5047\u8a2d\u6bcf\u500b\u97f3\u6846\u5305\u542b\uf9ba N \u7684\u6a23\u672c\u9ede\uff0c\u5247\u904e\uf9b2\uf961\u7684\u8a08\u7b97\u65b9\u5f0f\u5982 \u4e0b\u6240\u793a\u3002 Babble noise 1 (0dB\u30015dB\u300110dB) 16.67 % 46.67 % Babble noise 2 (0dB\u30015dB\u300110dB) 24.44 % 53.33 % \uf96b\u8003\u6587\u737b</td></tr><tr><td>Babble noise 3 (0dB\u30015dB\u300110dB)</td><td>25.56 %</td><td/><td/><td>66.67 %</td></tr><tr><td colspan=\"2\">sgn[x(n)] -sgn[x(n 61.11 % 48.89 % \u5716\u4e94\u3001\u8a9e\u97f3\u8fa8\uf9fc\u5668\uf9ca\u7a0b\u5716 0, x(n) if 1 -sgn[x(n)] | 1 -1 \uf0b3 Car noise 1 (0dB\u30015dB\u300110dB) sgn[x(n)] 2 1 ZCR N n \uf03d \uf03d \uf0e5 \uf03d Car noise 2 (0dB\u30015dB\u300110dB) Car noise 3 (0dB\u30015dB\u300110dB) 58.89 %</td><td>-1 | 1)] \uf03d</td><td>if</td><td>0 67.78 % x(n) \uf03c 66.67 % 72.22 %</td><td>(23)</td></tr></table>", |
| "num": null, |
| "type_str": "table" |
| } |
| } |
| } |
| } |
