| { | |
| "paper_id": "O15-3005", | |
| "header": { | |
| "generated_with": "S2ORC 1.0.0", | |
| "date_generated": "2023-01-19T08:09:59.010109Z" | |
| }, | |
| "title": "Investigating Modulation Spectrum Factorization Techniques for Robust Speech Recognition", | |
| "authors": [ | |
| { | |
| "first": "\u5f35\u5ead\u8c6a", | |
| "middle": [ | |
| "\uf02a" | |
| ], | |
| "last": "\u3001\u6d2a\u5b5d\u5b97", | |
| "suffix": "", | |
| "affiliation": {}, | |
| "email": "" | |
| }, | |
| { | |
| "first": "\uf02a", | |
| "middle": [], | |
| "last": "\u3001\u9673\u51a0\u5b87", | |
| "suffix": "", | |
| "affiliation": {}, | |
| "email": "" | |
| }, | |
| { | |
| "first": "\uf02b", | |
| "middle": [], | |
| "last": "\u3001\u738b\u65b0\u6c11", | |
| "suffix": "", | |
| "affiliation": {}, | |
| "email": "" | |
| }, | |
| { | |
| "first": "\uf02b", | |
| "middle": [], | |
| "last": "\u3001\u9673\u67cf\u7433", | |
| "suffix": "", | |
| "affiliation": {}, | |
| "email": "" | |
| }, | |
| { | |
| "first": "Ting-Hao", | |
| "middle": [], | |
| "last": "Chang", | |
| "suffix": "", | |
| "affiliation": {}, | |
| "email": "" | |
| }, | |
| { | |
| "first": "Hsiao-Tsung", | |
| "middle": [], | |
| "last": "Hung", | |
| "suffix": "", | |
| "affiliation": {}, | |
| "email": "" | |
| }, | |
| { | |
| "first": "Kuan-Yu", | |
| "middle": [], | |
| "last": "Chen", | |
| "suffix": "", | |
| "affiliation": {}, | |
| "email": "[email protected]" | |
| }, | |
| { | |
| "first": "Hsin-Min", | |
| "middle": [], | |
| "last": "Wang", | |
| "suffix": "", | |
| "affiliation": {}, | |
| "email": "" | |
| }, | |
| { | |
| "first": "Berlin", | |
| "middle": [], | |
| "last": "Chen", | |
| "suffix": "", | |
| "affiliation": {}, | |
| "email": "[email protected]" | |
| }, | |
| { | |
| "first": "\uf02a", | |
| "middle": [], | |
| "last": "\u570b\uf9f7\u81fa\u7063\u5e2b\u7bc4\u5927\u5b78\u8cc7\u8a0a\u5de5\u7a0b\u5b78\u7cfb", | |
| "suffix": "", | |
| "affiliation": {}, | |
| "email": "" | |
| } | |
| ], | |
| "year": "", | |
| "venue": null, | |
| "identifiers": {}, | |
| "abstract": "The performance of an automatic speech recognition (ASR) system often deteriorates sharply due to the interference from varying environmental noise. As such, the development of effective and efficient robustness techniques has long been a challenging research subject in the ASR community. In this article, we attempt to obtain noise-robust speech features through modulation spectrum processing of the original speech features. To this end, we explore the use of nonnegative matrix factorization (NMF) and its extensions on the magnitude modulation spectra of speech features so as to distill the most important and noise-resistant information cues that can benefit the ASR performance. The main contributions include three aspects: 1) we leverage the notion of sparseness to obtain more localized and parts-based representations of the magnitude modulation spectra with fewer basis vectors; 2) the prior knowledge of the similarities among training utterances is taken into account as an additional constraint during the NMF derivation; and 3) the resulting encoding vectors of NMF are further normalized so as to further enhance their robustness of representation. A series of experiments conducted on the Aurora-2 benchmark task demonstrate that our methods can deliver remarkable improvements over the baseline NMF method and achieve performance on par with or better than several widely-used robustness methods.", | |
| "pdf_parse": { | |
| "paper_id": "O15-3005", | |
| "_pdf_hash": "", | |
| "abstract": [ | |
| { | |
| "text": "The performance of an automatic speech recognition (ASR) system often deteriorates sharply due to the interference from varying environmental noise. As such, the development of effective and efficient robustness techniques has long been a challenging research subject in the ASR community. In this article, we attempt to obtain noise-robust speech features through modulation spectrum processing of the original speech features. To this end, we explore the use of nonnegative matrix factorization (NMF) and its extensions on the magnitude modulation spectra of speech features so as to distill the most important and noise-resistant information cues that can benefit the ASR performance. The main contributions include three aspects: 1) we leverage the notion of sparseness to obtain more localized and parts-based representations of the magnitude modulation spectra with fewer basis vectors; 2) the prior knowledge of the similarities among training utterances is taken into account as an additional constraint during the NMF derivation; and 3) the resulting encoding vectors of NMF are further normalized so as to further enhance their robustness of representation. A series of experiments conducted on the Aurora-2 benchmark task demonstrate that our methods can deliver remarkable improvements over the baseline NMF method and achieve performance on par with or better than several widely-used robustness methods.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Abstract", | |
| "sec_num": null | |
| } | |
| ], | |
| "body_text": [ | |
| { | |
| "text": ")\uff1a\u7b2c\u4e00\u7a2e\u985e\u578b\u70ba\u4ee5\u8072\u5b78\u6a21\u578b(Acoustic Model)\u70ba\u57fa\u790e\u4e4b\u5f37\u5065\u6027\u6280\u8853(Model-Based Techniques)\uff0c\u6b64\u985e\u65b9\u6cd5\u5927\u591a\u662f\u671f\u671b\u900f\u904e\u5c11\u91cf\u5728\u6e2c\u8a66\u74b0\u5883\u6240\u9304\u88fd\u7684\u8abf\u9069\u8a9e\u6599\u4f86\u5c0d\u8072\u5b78\u6a21\u578b \u8abf\u8b8a\u983b\u8b5c\u5206\u89e3\u6280\u8853\u65bc\u5f37\u5065\u8a9e\u97f3\u8fa8\u8b58\u4e4b\u7814\u7a76 89 \u9032\u884c\u8abf\u6574\uff0c\u4f7f\u8072\u5b78\u6a21\u578b\u53ef\u4ee5\u8fd1\u4f3c\u65bc\u8f38\u5165\u542b\u96dc\u8a0a\u8a9e\u97f3\u7684\u6a5f\u7387\u5206\u5e03\u53c3\u6578\uff0c\u9054\u5230\u964d\u4f4e\u74b0\u5883\u4e0d\u5339 \u914d \u6240 \u9020 \u6210 \u5f71 \u97ff \u7684 \u76ee \u7684 \u3002 \u7b2c \u4e8c \u985e \u662f \u4ee5 \u8a9e \u97f3 \u7279 \u5fb5 \u70ba \u57fa \u790e \u4e4b \u5f37 \u5065 \u6027 \u6280 \u8853 (Feature-Based Techniques)\u3002\u6b64\u985e\u65b9\u6cd5\u671f\u671b\u7d93\u904e\u9069\u7576\u7684\u6b63\u898f\u5316\u8655\u7406\u5f8c\uff0c\u80fd\u4f7f\u542b\u96dc\u8a0a\u8a9e\u97f3\u8207\u5176\u539f\u59cb\u4e7e\u6de8\u8a9e \u97f3\u8da8\u65bc\u4e00\u81f4\u3002\u6700\u5f8c\u7b2c\u4e09\u985e\u578b\u70ba\u7d9c\u5408\u5f0f\u5f37\u5065\u6027\u6280\u8853\uff0c\u5373\u540c\u6642\u5728\u7279\u5fb5\u8655\u7406\u548c\u6a21\u578b\u8a13\u7df4\u5169\u968e\u6bb5 \u505a\u6539\u5584\u3002 \u672c\u8ad6\u6587\u5c07\u63a2\u8a0e\u4ee5\u8a9e\u97f3\u7279\u5fb5\u70ba\u57fa\u790e\u4e4b\u5f37\u5065\u6027\u6280\u8853\u3002\u5176\u7814\u7a76\u7684\u8b70\u984c\u4e3b\u8981\u570d\u7e5e\u5728\u5c0d\u4f55\u7a2e\u7a7a \u9593\u6b63\u898f\u5316\uff1f\u4ee5\u53ca\u5728\u8a72\u7a7a\u9593\u61c9\u5982\u4f55\u6b63\u898f\u5316\uff1f\u5178\u578b\u65b9\u6cd5\u662f\u5c07\u6642\u9593\u5e8f\u5217\u57df(Temporal Domain)\u4e0a \u7684\u8a9e\u97f3\u7279\u5fb5\u8996\u70ba\u662f\u96a8\u6a5f\u8b8a\u6578(Random Variable)\u7684\u6a23\u672c(Samples)\uff0c\u5229\u7528\u89c0\u6e2c\u5230\u6a23\u672c\u53bb\u4f30\u6e2c \u96a8\u6a5f\u8b8a\u6578\u4e4b\u7d71\u8a08\u7279\u6027\uff0c\u9032\u800c\u5c0d\u8a9e\u97f3\u7279\u5fb5\u6642\u9593\u5e8f\u5217\u505a\u7dda\u6027\u6216\u975e\u7dda\u6027\u7684\u8f49\u63db\uff0c\u4f7f\u5176\u5728\u90e8\u5206\u6216 \u6574 \u9ad4 \u4e4b \u7d71 \u8a08 \u7279 \u6027 \u80fd \u7d93 \u904e \u6b63 \u898f \u5316 \u7684 \u8655 \u7406 \u3002 \u5e38 \u898b \u7684 \u65b9 \u6cd5 \u6709 \u7d71 \u8a08 \u5716 \u7b49 \u5316 \u6cd5 (Histogram Equalization, HEQ) (Torre et al., 2005) \u3001\u5012\u983b\u8b5c\u5e73\u5747\u503c\u6e1b\u53bb\u6cd5(Cepstral Mean Subtraction, CMS) (Furui, 1981) \u4ee5\u53ca\u5012\u983b\u8b5c\u5e73\u5747\u6578\u8207\u8b8a\u7570\u6578\u6b63\u898f\u5316\u6cd5(Cepstral Mean and Variance Normalization, CMVN) (Vikki & Laurila, 1998) (Sun et al., 2007) \u3001\u5206\u983b\u5f0f\u8abf\u8b8a\u983b\u8b5c\u7d71\u8a08\u6b63\u898f\u5316\u6cd5(Sub-Band Modulation Spectrum Compensation) (Huang et al., 2009) \u8207 \u5176 \u5b83 \u4e00 \u7cfb \u5217 \u8cc7 \u6599 \u5c0e \u5411 (Data-Driven)\u4e4b\u6642\u9593\u5e8f\u5217\u6ffe\u6ce2\u5668\u6cd5 (Xiao et al., 2008; Hermansky & Morgan, 1994) (Kanedera et al., 1997) \u3002\u540c\u6642\uff0c\u8abf\u8b8a\u983b\u8b5c\u4e4b\u4f4e\u983b\u6210\u5206(\u7d04 1Hz \u81f3 16Hz)\u5c0d\u65bc\u8a9e\u97f3\u8fa8\u8b58\u6b63\u78ba\u7387\u4e5f\u6709\u5bc6\u5207\u7684\u95dc\u4fc2\uff0c\u6f5b\u85cf\u6709\u91cd\u8981\u7684\u8a9e\u610f\u8cc7\u8a0a\u3002\u5176\u4e2d\uff0c\u6700\u91cd\u8981 \u7684\u662f\u4f4d\u65bc 4 Hz \u9644\u8fd1\uff0c\u6709\u5b78\u8005\u6307\u51fa\uff0c4 Hz \u662f\u4eba\u8033\u807d\u89ba\u6700\u70ba\u654f\u611f\u4e4b\u8abf\u8b8a\u983b\u7387 (Hermansky, 1998) \uff1b \u53e6\u6709\u5b78\u8005\u4e5f\u8a8d\u70ba\uff0c4 Hz \u70ba\u4eba\u985e\u5927\u8166\u76ae\u5c64\u611f\u77e5\u4e4b\u91cd\u8981\u8abf\u8b8a\u983b\u7387 (Greenberg, 1997) ", | |
| "cite_spans": [ | |
| { | |
| "start": 582, | |
| "end": 602, | |
| "text": "(Torre et al., 2005)", | |
| "ref_id": "BIBREF17" | |
| }, | |
| { | |
| "start": 646, | |
| "end": 659, | |
| "text": "(Furui, 1981)", | |
| "ref_id": "BIBREF2" | |
| }, | |
| { | |
| "start": 725, | |
| "end": 748, | |
| "text": "(Vikki & Laurila, 1998)", | |
| "ref_id": "BIBREF18" | |
| }, | |
| { | |
| "start": 749, | |
| "end": 767, | |
| "text": "(Sun et al., 2007)", | |
| "ref_id": "BIBREF16" | |
| }, | |
| { | |
| "start": 826, | |
| "end": 846, | |
| "text": "(Huang et al., 2009)", | |
| "ref_id": "BIBREF8" | |
| }, | |
| { | |
| "start": 890, | |
| "end": 909, | |
| "text": "(Xiao et al., 2008;", | |
| "ref_id": "BIBREF19" | |
| }, | |
| { | |
| "start": 910, | |
| "end": 935, | |
| "text": "Hermansky & Morgan, 1994)", | |
| "ref_id": "BIBREF5" | |
| }, | |
| { | |
| "start": 936, | |
| "end": 959, | |
| "text": "(Kanedera et al., 1997)", | |
| "ref_id": "BIBREF9" | |
| }, | |
| { | |
| "start": 1061, | |
| "end": 1078, | |
| "text": "(Hermansky, 1998)", | |
| "ref_id": "BIBREF6" | |
| }, | |
| { | |
| "start": 1111, | |
| "end": 1128, | |
| "text": "(Greenberg, 1997)", | |
| "ref_id": "BIBREF3" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "EQUATION", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [ | |
| { | |
| "start": 0, | |
| "end": 8, | |
| "text": "EQUATION", | |
| "ref_id": "EQREF", | |
| "raw_str": "V WH W H (5) \u5176\u4e2dV \u2208 \u70ba\u4e00\u500b\u975e\u8ca0\u77e9\u9663\uff0c\u800c\u5169\u500b\u88ab\u5206\u89e3\u51fa\u4f86\u7684\u975e\u8ca0\u77e9\u9663\u5206\u5225\u70baW \u2208 \u548c H \u2208 \uff0c\u5982\u5716 1 \u6240\u793a\u3002 \u5716 1. \u975e\u8ca0\u77e9\u9663\u5206\u89e3\u6cd5(NMF)\u793a\u610f\u5716 \u5176\u4e2d\u77e9\u9663 W \u6240\u5305\u7684 K \u884c\u5373\u70ba\u57fa\u5e95\u5411\u91cf\uff0c\u77e9\u9663 H \u4e2d\u7684\u6bcf\u4e00\u884c\u5247\u901a\u5e38\u88ab\u7a31\u70ba\u7de8\u78bc\u5411\u91cf (Encoding)\uff0c\u6709\u8457\u6b0a\u91cd\u7684\u6982\u5ff5\uff0c\u8207\u57fa\u5e95\u5411\u91cf\u9032\u884c\u7dda\u6027\u7d44\u5408\u53bb\u8fd1\u4f3c\u8cc7\u6599\u77e9\u9663 V\u3002I \u662f\u6bcf\u7b46\u8cc7 \u6599\u5411\u91cf\u7684\u7dad\u5ea6\u5927\u5c0f\uff1bJ \u70ba\u6240\u6709\u8cc7\u6599\u5411\u91cf\u7684\u500b\u6578\uff1bK \u70ba\u57fa\u5e95\u5411\u91cf\u7684\u6578\u91cf\u3002\u53c3\u6578 K \u662f\u53ef\u4ee5\u81ea \u884c\u6c7a\u5b9a\u7684\uff0c\u901a\u5e38\u6703\u9078\u64c7\u5c0f\u65bc I \u8207 J\uff0c\u4f46\u9084\u662f\u6703\u6709\u9078\u64c7\u7684\u9650\u5236\uff1a (6) \u5f0f(6)\u662f\u5b78\u8005\u904e\u53bb\u6240\u63d0\u51fa\u66f4\u78ba\u5207\u7684\u57fa\u5e95\u5411\u91cf\u500b\u6578\u9078\u64c7\u9650\u5236\u95dc\u4fc2\u5f0f\u3002\u5728\u975e\u8ca0\u77e9\u9663\u5206\u89e3\u6cd5\u7684\u65b9 \u6cd5\u4e2d\uff0c\u6709\u8457\u8cc7\u6599\u58d3\u7e2e\u7684\u6982\u5ff5\uff0c\u82e5\u662f K \u7684\u6578\u76ee\u9078\u64c7\u5f97\u8d8a\u5c11\uff0c\u4ee3\u8868\u58d3\u7e2e\u7684\u6bd4\u7387\u8d8a\u9ad8\u3002\u56e0\u70ba\u6211 \u5011\u5c0d\u8cc7\u6599\u9032\u884c\u4e86\u58d3\u7e2e\u7684\u52d5\u4f5c\uff0c\u6240\u4ee5\u58d3\u7e2e\u5f8c\u7684\u8cc7\u6599\u8ddf\u539f\u59cb\u7684\u8cc7\u6599\u4f86\u6bd4\u8f03\u5fc5\u5b9a\u6703\u6709\u4e00\u4e9b\u8cc7\u6599 \u662f\u5728\u58d3\u7e2e\u904e\u7a0b\u4e2d\u88ab\u907a\u5931\u4e86\u3002\u6211\u5011\u5e0c\u671b\u907a\u5931\u7684\u90e8\u5206\u8cc7\u6599\u8d8a\u5c11\u8d8a\u597d\uff0c\u6240\u4ee5\u53ef\u4ee5\u5b9a\u7fa9\u6e1b\u640d\u51fd\u6578 (Loss Function)\u4f86\u6e2c\u91cf\u8cc7\u6599\u524d\u5f8c\u7684\u76f8\u4f3c\u5ea6\u3002\u6e2c\u91cf\u7531\u5169\u500b\u56e0\u5b50\u77e9\u9663 W \u8207 H \u6240\u91cd\u5efa\u7684\u8a0a\u865f\u039b \u8207\u539f\u59cb\u8a0a\u865f V \u4e4b\u9593\u7684\u8ddd\u96e2\uff0c\u5c0d\u5206\u89e3\u7d50\u679c\u8207\u539f\u59cb\u8cc7\u6599\u7684\u8fd1\u4f3c\u7a0b\u5ea6\u4f5c\u91cf\u5316(Quantification)\u3002 \u8abf\u8b8a\u983b\u8b5c\u5206\u89e3\u6280\u8853\u65bc\u5f37\u5065\u8a9e\u97f3\u8fa8\u8b58\u4e4b\u7814\u7a76 93 \u975e\u8ca0\u77e9\u9663\u5206\u89e3\u6cd5\u5e38\u898b\u7684\u6e1b\u640d\u51fd\u6578\u70ba\u6b50\u6c0f\u8ddd\u96e2(Euclidian Distance \u6216 Frobenius Norm)\uff1a D V||WH ||V WH|| V WH ,", | |
| "eq_num": "(7)" | |
| } | |
| ], | |
| "section": "", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "EQUATION", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [ | |
| { | |
| "start": 0, | |
| "end": 8, | |
| "text": "EQUATION", | |
| "ref_id": "EQREF", | |
| "raw_str": "D V||WH \u662f\u85c9\u7531\u6b50\u6c0f\u8ddd\u96e2\u6240\u63d0\u51fa\u7684\u6e1b\u640d\u51fd\u6578\u3002\u7576\u91cd\u5efa\u8a0a\u865f\u039b\u8207\u539f\u59cb\u4fe1\u865f V \u76f8\u7b49\u6642\uff0c\u5247 D V||WH 0\u3002\u53e6\u4e00\u500b\u6e1b\u640d\u51fd\u6578\u5247\u662f\u57fa\u65bc KL \u6563\u5ea6(Kullback-Leibler Divergence)\uff1a D V||WH V ln V WH V WH ,", | |
| "eq_num": "(8)" | |
| } | |
| ], | |
| "section": "", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "\u7576\u539f\u59cb\u4fe1\u865f V", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "EQUATION", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [ | |
| { | |
| "start": 0, | |
| "end": 8, | |
| "text": "EQUATION", | |
| "ref_id": "EQREF", | |
| "raw_str": "H \u2190 H \u2211 WS V WS H \u2044 \u2211 W \u2190 W \u2211 SH V W SH \u2044 \u2211 SH 15 \u800c\u5176\u5b83\u90e8\u5206\u7684\u6f14\u7b97\u6cd5\u6d41\u7a0b\u540c\u50b3\u7d71\u975e\u8ca0\u77e9\u9663\u5206\u89e3\u6cd5\u3002 3.3 \u57fa\u65bc\u5716\u6b63\u5247\u5316\u975e\u8ca0\u77e9\u9663\u5206\u89e3\u6cd5(GNMF) \u57fa \u65bc \u5716 \u6b63 \u5247 \u5316 \u975e \u8ca0 \u77e9 \u9663 \u5206 \u89e3 \u6cd5 (Graph Regularized Non-negative Matrix Factorization, GNMF)(Cai et al., 2011)\u7684\u4e3b\u8981\u76ee\u7684\u5728\u65bc\u4fdd\u7559\u8cc7\u6599\u7684\u5c40\u90e8\u4e0d\u8b8a\u6027(Locally Invariant)(Hadsell et al., 2006)\uff0c\u610f\u6307\u539f\u672c\u76f8\u9130\u7684\u8cc7\u6599\u5411\u91cf\u7d93\u904e\u964d\u7dad\u6216\u6295\u5f71\u5f8c\u4ecd\u7136\u7dad\u6301\u76f8\u9130\u8fd1\u3002\u8cc7\u6599\u5411\u91cf\u9593 \u7684\u9060\u8fd1\u95dc\u4fc2\uff0c\u6216\u5e7e\u4f55\u7d50\u69cb\u8cc7\u8a0a\u53ef\u4ee5\u7528\u3127\u6b0a\u91cd\u77e9\u9663 E \u8868\u793a\uff0c\u5176\u7dad\u5ea6\u662f\u7b49\u65bc\u8cc7\u6599\u5411\u91cf\u6578\u91cf\u6240 \u5f62\u6210\u7684\u65b9\u9663\u3002\u6700\u5f8c\u5c07\u6b0a\u91cd\u77e9\u9663 E \u7d0d\u5165\u6e1b\u640d\u51fd\u5f0f\u4e2d\uff0c\u505a\u70ba\u7de8\u78bc\u77e9\u9663\u7684\u6b63\u5247\u9805(Regularization Term)\u3002 \u4ee4 , \u2026 , \u70ba\u7de8\u78bc\u77e9\u9663H\u7684\u7b2c j \u884c\uff0c \u53ef\u88ab\u8996\u70ba\u662f\u7b2c \u500b\u8cc7\u6599\u5411\u91cf\u76f8\u5c0d\u65bc\u65b0 \u7684\u57fa\u5e95\u77e9\u9663W\u4e4b\u65b0\u8868\u793a\u3002\u5728\u6b64\u6211\u5011\u8a0e\u8ad6\u8f03\u5e38\u898b\u7684\u6b50\u5f0f\u8ddd\u96e2\uff1a ,", | |
| "eq_num": "(16)" | |
| } | |
| ], | |
| "section": "", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "EQUATION", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [ | |
| { | |
| "start": 0, | |
| "end": 8, | |
| "text": "EQUATION", | |
| "ref_id": "EQREF", | |
| "raw_str": "\u6b64\u8ddd\u96e2\u7528\u4f86\u6e2c\u91cf\u76f8\u5c0d\u65bc\u65b0\u7684\u57fa\u5e95\u77e9\u9663W\uff0c\u800c\u5169\u500b\u8cc7\u6599\u5411\u91cf \u8207 \u5728\u4f4e\u7dad\u5ea6\u7a7a\u9593\u4e2d\u8868\u793a\u4e4b \u9593\u7684\u5dee\u7570(Dissimilarity)\uff0c\u8ddd\u96e2\u51fd\u5f0f\u503c\u8d8a\u5927\u4ee3\u8868\u6b64\u5169\u500b\u8cc7\u6599\u5411\u91cf \u8207 \u5f7c\u6b64\u5dee\u7570\u8d8a\u5927\u3002 1 2 , E D E ,", | |
| "eq_num": "(17)" | |
| } | |
| ], | |
| "section": "", | |
| "sec_num": null | |
| } | |
| ], | |
| "back_matter": [], | |
| "bib_entries": { | |
| "BIBREF0": { | |
| "ref_id": "b0", | |
| "title": "Graph regularized nonnegative matrix factorization for data representation", | |
| "authors": [ | |
| { | |
| "first": "D", | |
| "middle": [], | |
| "last": "Cai", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "X", | |
| "middle": [], | |
| "last": "He", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "J", | |
| "middle": [], | |
| "last": "Han", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "T", | |
| "middle": [ | |
| "S" | |
| ], | |
| "last": "Huang", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2011, | |
| "venue": "IEEE Transactions on Pattern Analysis and Machine Intelligence", | |
| "volume": "33", | |
| "issue": "8", | |
| "pages": "1548--1560", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Cai, D., He, X., Han, J., & Huang, T. S. (2011). Graph regularized nonnegative matrix factorization for data representation. IEEE Transactions on Pattern Analysis and Machine Intelligence, 33(8), 1548-1560.", | |
| "links": null | |
| }, | |
| "BIBREF1": { | |
| "ref_id": "b1", | |
| "title": "Modulation spectrum factorization for robust speech recognition", | |
| "authors": [ | |
| { | |
| "first": "W.-Y", | |
| "middle": [], | |
| "last": "Chu", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "J.-W", | |
| "middle": [], | |
| "last": "Hung", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "B", | |
| "middle": [], | |
| "last": "Chen", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2011, | |
| "venue": "Proceedings of the APSIPA Annual Summit and Conference", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "18--21", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Chu, W.-Y., Hung, J.-W., & Chen, B. (2011). Modulation spectrum factorization for robust speech recognition. In Proceedings of the APSIPA Annual Summit and Conference, 18-21.", | |
| "links": null | |
| }, | |
| "BIBREF2": { | |
| "ref_id": "b2", | |
| "title": "Cepstral analysis techniques for automatic speaker verification", | |
| "authors": [ | |
| { | |
| "first": "S", | |
| "middle": [], | |
| "last": "Furui", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1981, | |
| "venue": "IEEE Transactions on Acoustic, Speech and Signal Processing", | |
| "volume": "29", | |
| "issue": "2", | |
| "pages": "254--272", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Furui, S. (1981). Cepstral analysis techniques for automatic speaker verification. IEEE Transactions on Acoustic, Speech and Signal Processing, 29(2), 254-272.", | |
| "links": null | |
| }, | |
| "BIBREF3": { | |
| "ref_id": "b3", | |
| "title": "On the origins of speech intelligibility in the real world", | |
| "authors": [ | |
| { | |
| "first": "S", | |
| "middle": [], | |
| "last": "Greenberg", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1997, | |
| "venue": "Proceedings of the ESCA-NATO Tutorial and Research Workshop on Robust Speech Recognition for Unknown Communication Channels", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Greenberg, S. (1997). On the origins of speech intelligibility in the real world. In Proceedings of the ESCA-NATO Tutorial and Research Workshop on Robust Speech Recognition for Unknown Communication Channels.", | |
| "links": null | |
| }, | |
| "BIBREF4": { | |
| "ref_id": "b4", | |
| "title": "Dimensionality reduction by learning an invariant mapping", | |
| "authors": [ | |
| { | |
| "first": "R", | |
| "middle": [], | |
| "last": "Hadsell", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "S", | |
| "middle": [], | |
| "last": "Chopra", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Y", | |
| "middle": [], | |
| "last": "Lecun", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2006, | |
| "venue": "Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "1735--1742", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Hadsell, R., Chopra, S., & LeCun Y. (2006). Dimensionality reduction by learning an invariant mapping. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 1735-1742.", | |
| "links": null | |
| }, | |
| "BIBREF5": { | |
| "ref_id": "b5", | |
| "title": "RASTA processing of speech", | |
| "authors": [ | |
| { | |
| "first": "H", | |
| "middle": [], | |
| "last": "Hermansky", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "N", | |
| "middle": [], | |
| "last": "Morgan", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1994, | |
| "venue": "IEEE Transactions on Speech and Audio Processing", | |
| "volume": "2", | |
| "issue": "4", | |
| "pages": "578--589", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Hermansky, H., & Morgan, N. (1994). RASTA processing of speech. IEEE Transactions on Speech and Audio Processing, 2(4), 578-589.", | |
| "links": null | |
| }, | |
| "BIBREF6": { | |
| "ref_id": "b6", | |
| "title": "Should Recognizers Have Ears? Speech Communication", | |
| "authors": [ | |
| { | |
| "first": "H", | |
| "middle": [], | |
| "last": "Hermansky", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1998, | |
| "venue": "", | |
| "volume": "25", | |
| "issue": "", | |
| "pages": "3--27", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Hermansky, H. (1998). Should Recognizers Have Ears? Speech Communication, 25(1-3), 3-27.", | |
| "links": null | |
| }, | |
| "BIBREF7": { | |
| "ref_id": "b7", | |
| "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 the ISCA ITRW ASR", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Hirsch, H. G., & Pearce, D. (2000). The AURORA experimental framework for the performance evaluations of speech recognition systems under noisy conditions. In Proceedings of the ISCA ITRW ASR.", | |
| "links": null | |
| }, | |
| "BIBREF8": { | |
| "ref_id": "b8", | |
| "title": "A study of sub-band modulation spectrum compensation for robust speech recognition", | |
| "authors": [ | |
| { | |
| "first": "S.-Y", | |
| "middle": [], | |
| "last": "Huang", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "W.-H", | |
| "middle": [], | |
| "last": "Tu", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "J.-W", | |
| "middle": [], | |
| "last": "Hung", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2009, | |
| "venue": "Proceedings of the ROCLING XXI: Conference on Computational Linguistics and Speech Processing", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Huang, S.-Y., Tu, W.-H., & Hung, J.-W. (2009). A study of sub-band modulation spectrum compensation for robust speech recognition. In Proceedings of the ROCLING XXI: Conference on Computational Linguistics and Speech Processing.", | |
| "links": null | |
| }, | |
| "BIBREF9": { | |
| "ref_id": "b9", | |
| "title": "On the importance of various modulation frequencies for speech recognition", | |
| "authors": [ | |
| { | |
| "first": "N", | |
| "middle": [], | |
| "last": "Kanedera", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "T", | |
| "middle": [], | |
| "last": "Arai", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "H", | |
| "middle": [], | |
| "last": "Hermansky", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "M", | |
| "middle": [], | |
| "last": "Pavel", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1997, | |
| "venue": "Proceedings of the European Conference on Speech Communication and Technology", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Kanedera, N., Arai, T., Hermansky, H., & Pavel, M. (1997). On the importance of various modulation frequencies for speech recognition. In Proceedings of the European Conference on Speech Communication and Technology.", | |
| "links": null | |
| }, | |
| "BIBREF10": { | |
| "ref_id": "b10", | |
| "title": "Effective modulation spectrum factorization for robust speech recognition", | |
| "authors": [ | |
| { | |
| "first": "Y.-C", | |
| "middle": [], | |
| "last": "Kao", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Y.-T", | |
| "middle": [], | |
| "last": "Wang", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "B", | |
| "middle": [], | |
| "last": "Chen", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2014, | |
| "venue": "Proceedings of the Annual Conference of the International Speech Communication Association", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "2724--2728", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Kao, Y.-C., Wang, Y.-T., & Chen, B. (2014). Effective modulation spectrum factorization for robust speech recognition. In Proceedings of the Annual Conference of the International Speech Communication Association, 2724-2728.", | |
| "links": null | |
| }, | |
| "BIBREF11": { | |
| "ref_id": "b11", | |
| "title": "Learning the parts of objects by non-negative matrix factorization", | |
| "authors": [ | |
| { | |
| "first": "D", | |
| "middle": [ | |
| "D" | |
| ], | |
| "last": "Lee", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "H", | |
| "middle": [ | |
| "S" | |
| ], | |
| "last": "Seung", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1999, | |
| "venue": "Nature", | |
| "volume": "401", | |
| "issue": "", | |
| "pages": "788--791", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Lee, D. D., & Seung, H. S. (1999). Learning the parts of objects by non-negative matrix factorization. Nature, 401, 788-791.", | |
| "links": null | |
| }, | |
| "BIBREF12": { | |
| "ref_id": "b12", | |
| "title": "Algorithms for Non-negative Matrix Factorization", | |
| "authors": [ | |
| { | |
| "first": "D", | |
| "middle": [ | |
| "D" | |
| ], | |
| "last": "Lee", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "H", | |
| "middle": [ | |
| "S" | |
| ], | |
| "last": "Seung", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2000, | |
| "venue": "Proceedings of the Annual Conference on Neural Information Processing Systems", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "556--562", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Lee, D. D., & Seung, H. S. (2000). Algorithms for Non-negative Matrix Factorization. In Proceedings of the Annual Conference on Neural Information Processing Systems, 556-562.", | |
| "links": null | |
| }, | |
| "BIBREF13": { | |
| "ref_id": "b13", | |
| "title": "Exploring the use of speech features and their corresponding distribution characteristics for robust speech recognition", | |
| "authors": [ | |
| { | |
| "first": "S.-H", | |
| "middle": [], | |
| "last": "Lin", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "B", | |
| "middle": [], | |
| "last": "Chen", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Y.-M", | |
| "middle": [], | |
| "last": "Yeh", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2009, | |
| "venue": "IEEE Transactions on Audio, Speech and Language Processing", | |
| "volume": "17", | |
| "issue": "1", | |
| "pages": "84--94", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Lin, S.-H., Chen, B., & Yeh, Y.-M. (2009). Exploring the use of speech features and their corresponding distribution characteristics for robust speech recognition. IEEE Transactions on Audio, Speech and Language Processing, 17(1), 84-94.", | |
| "links": null | |
| }, | |
| "BIBREF14": { | |
| "ref_id": "b14", | |
| "title": "Evaluation of a noise-robust DSR front-end on Aurora databases", | |
| "authors": [ | |
| { | |
| "first": "D", | |
| "middle": [], | |
| "last": "Macho", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "L", | |
| "middle": [], | |
| "last": "Mauuary", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "B", | |
| "middle": [], | |
| "last": "No\u00e9", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Y", | |
| "middle": [ | |
| "M" | |
| ], | |
| "last": "Cheng", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "D", | |
| "middle": [], | |
| "last": "Ealey", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "D", | |
| "middle": [], | |
| "last": "Jouvet", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "H", | |
| "middle": [], | |
| "last": "Kelleher", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "D", | |
| "middle": [], | |
| "last": "Pearce", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "F", | |
| "middle": [], | |
| "last": "Saadoun", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2002, | |
| "venue": "Proceedings of the Annual Conference of the International Speech Communication Association", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Macho, D., Mauuary, L., No\u00e9, B., Cheng, Y. M., Ealey, D., Jouvet, D., Kelleher, H., Pearce, D., & Saadoun, F. (2002). Evaluation of a noise-robust DSR front-end on Aurora databases. In Proceedings of the Annual Conference of the International Speech Communication Association.", | |
| "links": null | |
| }, | |
| "BIBREF15": { | |
| "ref_id": "b15", | |
| "title": "Nonsmooth nonnegtive matrix facotorization (nsNMF)", | |
| "authors": [ | |
| { | |
| "first": "A", | |
| "middle": [], | |
| "last": "Pascual-Montano", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "J", | |
| "middle": [ | |
| "M" | |
| ], | |
| "last": "Carazo", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "K", | |
| "middle": [], | |
| "last": "Kochi", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "D", | |
| "middle": [], | |
| "last": "Lehmann", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "R", | |
| "middle": [ | |
| "D" | |
| ], | |
| "last": "Marqui", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2006, | |
| "venue": "IEEE Transactions on Pattern Analysis and Machine Intelligence", | |
| "volume": "28", | |
| "issue": "3", | |
| "pages": "403--415", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Pascual-Montano, A., Carazo, J. M., Kochi, K., Lehmann, D., & Pascual-Marqui, R. D. (2006). Nonsmooth nonnegtive matrix facotorization (nsNMF). IEEE Transactions on Pattern Analysis and Machine Intelligence, 28(3), 403-415.", | |
| "links": null | |
| }, | |
| "BIBREF16": { | |
| "ref_id": "b16", | |
| "title": "Modulation Spectrum Equalization for robust Speech Recognition", | |
| "authors": [ | |
| { | |
| "first": "L.-C", | |
| "middle": [], | |
| "last": "Sun", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "C.-W", | |
| "middle": [], | |
| "last": "Hsu", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "L.-S", | |
| "middle": [], | |
| "last": "Lee", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2007, | |
| "venue": "Proceedings of IEEE Workshop on Automatic Speech Recognition and Understanding", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Sun, L.-C., Hsu, C.-W., & Lee, L.-S. (2007). Modulation Spectrum Equalization for robust Speech Recognition. In Proceedings of IEEE Workshop on Automatic Speech Recognition and Understanding.", | |
| "links": null | |
| }, | |
| "BIBREF17": { | |
| "ref_id": "b17", | |
| "title": "Histogram equalization of speech representation for robust speech recognition", | |
| "authors": [ | |
| { | |
| "first": "A", | |
| "middle": [ | |
| "D L" | |
| ], | |
| "last": "Torre", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "A", | |
| "middle": [ | |
| "M J" | |
| ], | |
| "last": "Peinado", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "C", | |
| "middle": [], | |
| "last": "Segura", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "J", | |
| "middle": [ | |
| "L" | |
| ], | |
| "last": "Perez-Cordoba", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "M", | |
| "middle": [ | |
| "C" | |
| ], | |
| "last": "Benitez", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "A", | |
| "middle": [ | |
| "J" | |
| ], | |
| "last": "Rubio", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2005, | |
| "venue": "IEEE Transactions on Speech and Audio Processing", | |
| "volume": "13", | |
| "issue": "3", | |
| "pages": "355--366", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Torre, A. D. L., Peinado, A. M. J., Segura, C., Perez-Cordoba, J. L., Benitez, M. C., & Rubio, A. J. (2005). Histogram equalization of speech representation for robust speech recognition. IEEE Transactions on Speech and Audio Processing, 13(3), 355-366.", | |
| "links": null | |
| }, | |
| "BIBREF18": { | |
| "ref_id": "b18", | |
| "title": "Segmental feature vector normalization for noise robust speech recognition", | |
| "authors": [ | |
| { | |
| "first": "A", | |
| "middle": [], | |
| "last": "Vikki", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "K", | |
| "middle": [], | |
| "last": "Laurila", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1998, | |
| "venue": "Speech Communication", | |
| "volume": "25", | |
| "issue": "", | |
| "pages": "133--147", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Vikki, A., & Laurila, K. (1998), Segmental feature vector normalization for noise robust speech recognition. Speech Communication, 25, 133-147.", | |
| "links": null | |
| }, | |
| "BIBREF19": { | |
| "ref_id": "b19", | |
| "title": "Normalization of the speech modulation spectra for robust speech recognition", | |
| "authors": [ | |
| { | |
| "first": "X", | |
| "middle": [], | |
| "last": "Xiao", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "E", | |
| "middle": [ | |
| "S" | |
| ], | |
| "last": "Chng", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "H", | |
| "middle": [], | |
| "last": "Li", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2008, | |
| "venue": "IEEE Transactions on Speech and Audio Processing", | |
| "volume": "16", | |
| "issue": "8", | |
| "pages": "1662--1674", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Xiao, X., Chng, E. S., & Li, H. (2008). Normalization of the speech modulation spectra for robust speech recognition. IEEE Transactions on Speech and Audio Processing, 16(8), 1662-1674.", | |
| "links": null | |
| } | |
| }, | |
| "ref_entries": { | |
| "TABREF3": { | |
| "html": null, | |
| "text": "", | |
| "type_str": "table", | |
| "content": "<table><tr><td/><td colspan=\"2\">\u8abf\u8b8a\u983b\u8b5c\u5206\u89e3\u6280\u8853\u65bc\u5f37\u5065\u8a9e\u97f3\u8fa8\u8b58\u4e4b\u7814\u7a76</td><td>91 \u5f35\u5ead\u8c6a \u7b49</td></tr><tr><td colspan=\"4\">\u53ef\u4ee5\u540c\u6642\u5c0d\u5176\u5e73\u5747\u503c\u548c\u6a19\u6e96\u5dee\u4f86\u9032\u884c\u6b63\u898f\u5316\uff1a \u80fd\u5920\u63d0\u4f9b\u975e\u8ca0\u7684\u57fa\u5e95\u5411\u91cf(Nonnegative Basis Vectors)\uff0c\u4e14\u4e5f\u80fd\u5920\u64c1\u6709\u4fdd\u8b49\u7531\u57fa\u5e95\u5411\u91cf\u7d44</td></tr><tr><td colspan=\"4\">| \u5408\u800c\u6210\u4e4b\u8cc7\u6599\u4e5f\u70ba\u975e\u8ca0\u7684\u7279\u6027\u3002\u975e\u8ca0\u77e9\u9663\u5206\u89e3\u6cd5\u7684\u53e6\u4e00\u500b\u91cd\u8981\u7279\u6027\u662f\u60f3\u8981\u5b78\u7fd2\u4ee5\u90e8\u5206\u70ba | (3) \u57fa\u790e(Parts-Based)\u4e4b\u7dda\u6027\u8868\u793a\u6cd5\u4f86\u8868\u793a\u539f\u59cb\u7684\u8cc7\u6599\uff0c\u4e14\u6b64\u7dda\u6027\u8868\u793a\u6cd5\u662f\u4e00\u500b\u52a0\u6cd5\u7684\u7d44\u5408\u6a21</td></tr><tr><td colspan=\"4\">\u5f0f\u3002\u9019\u7a2e\u4ee5\u90e8\u5206\u70ba\u57fa\u790e\u7684\u6982\u5ff5\u65b9\u6cd5\u64c1\u6709\u76f4\u89c0\u7684\u6027\u8cea\uff0c\u800c\u4e14\u5c0d\u65bc\u4e00\u500b\u7279\u5b9a\u4efb\u52d9\u4f86\u8aaa\uff0c\u5728\u8207</td></tr><tr><td colspan=\"4\">\u5728\u5f0f(3)\u4e2d\uff0c \u8207 \u70ba\u55ae\u4e00\u8a9e\u53e5\u7684\u8abf\u8b8a\u983b\u8b5c\u5f37\u5ea6\u6210\u5206\u4e4b\u5e73\u5747\u503c\u8207\u6a19\u6e96\u5dee\uff1b \u8207 \u70ba\u6240\u6709\u8a13 \u5176\u5b83\u5206\u89e3\u65b9\u6cd5\u76f8\u6bd4\u4e0b\u53ef\u4ee5\u5f97\u5230\u6bd4\u8f03\u9ad8\u7684\u89e3\u91cb\u6027\u3002\u904e\u53bb\u6709\u5b78\u8005\u61c9\u7528\u975e\u8ca0\u77e9\u9663\u5206\u89e3\u6cd5\u5728\u5f71\u50cf</td></tr><tr><td colspan=\"4\">\u7df4\u8a9e\u53e5\u7684\u8abf\u8b8a\u983b\u8b5c\u5f37\u5ea6\u6210\u5206\u4e4b\u5e73\u5747\u503c\u8207\u6a19\u6e96\u5dee\uff0c \u8655\u7406\u7684\u9818\u57df\uff0c\u4f8b\u5982\u4eba\u81c9\u5f71\u6a23\u53ef\u4ee5\u7528\u70ba\u4e94\u5b98\u7b49\u5c40\u90e8\u5f71\u50cf\u505a\u70ba\u975e\u8ca0\u57fa\u5e95\u5411\u91cf\u7d93\u7531\u7dda\u6027\u7d44\u5408(\u7dda \u4fbf\u662f\u66f4\u65b0\u904e\u5f8c\u7684\u8abf\u8b8a\u983b\u8b5c\u5f37\u5ea6\u6210\u5206\u3002</td></tr><tr><td colspan=\"4\">\u6027\u7de8\u78bc)\u800c\u7522\u751f\u3002\u82e5\u662f\u4f7f\u7528\u4e0a\u8ff0\u6240\u63d0\u5230\u7684\uff0c\u4f8b\u5982 PCA\uff0c\u5728\u5206\u89e3\u8209\u8b49\u7522\u751f\u57fa\u5e95\u5411\u91cf\u7684\u904e\u7a0b\u4e2d</td></tr><tr><td colspan=\"4\">2.4 \u8abf\u8b8a\u983b\u8b5c\u7d71\u8a08\u5716\u7b49\u5316\u6cd5(Spectral Histogram Equalization, SHE) \u53ef\u80fd\u6703\u7522\u751f\u8ca0\u503c\uff0c\u9019\u4e9b\u8ca0\u503c\u5728\u5f71\u50cf\u8655\u7406\u7576\u4e2d\u6703\u96e3\u4ee5\u89e3\u91cb\u3002\u800c\u5728\u8a9e\u97f3\u9818\u57df\u65b9\u9762\uff0c\u8a9e\u97f3\u7684\u7279</td></tr><tr><td colspan=\"4\">\u5229\u7528\u975e\u7dda\u6027\u7684\u8f49\u63db(Nonlinear Transformation)\uff0c\u4e0d\u50c5\u5c07\u8abf\u8b8a\u983b\u8b5c\u5f37\u5ea6\u6210\u5206\u4e4b\u5e73\u5747\u503c\u8207\u6a19\u6e96 \u5fb5\u503c\u6709\u6b63\u6709\u8ca0\uff0c\u6240\u4ee5\u8f03\u96e3\u4ee5\u76f4\u63a5\u5730\u4f7f\u7528\u975e\u8ca0\u77e9\u9663\u5206\u89e3\u6cd5\uff1b\u76f4\u5230\u8fd1\u671f\u6709\u5b78\u8005\u5c07\u975e\u8ca0\u77e9\u9663\u5206</td></tr><tr><td colspan=\"4\">\u5dee(\u6216\u8b8a\u7570\u6578)\u4f5c\u6b63\u898f\u5316\uff0c\u800c\u662f\u6574\u9ad4\u4e0a\u4f7f\u5f97\u8a13\u7df4\u8a9e\u53e5\u8207\u6e2c\u8a66\u8a9e\u53e5\u7684\u8abf\u8b8a\u983b\u8b5c\u5f37\u5ea6\u6210\u5206\u8da8\u65bc \u89e3\u6cd5\u7528\u5728\u5206\u6790\u8abf\u8b8a\u983b\u8b5c\u5f37\u5ea6\u4ee5\u64f7\u53d6\u91cd\u8981\u8a9e\u97f3\u7279\u5fb5(Chu et al., 2011)\uff0c\u800c\u53ef\u4ee5\u5f97\u5230\u4e86\u4e0d\u932f\u7684</td></tr><tr><td colspan=\"4\">\u64c1\u6709\u540c\u4e00\u500b\u6a5f\u7387\u5206\u5e03\u51fd\u6578\uff0c\u6b63\u898f\u5316\u5168\u90e8\u968e\u5c64\u7684\u52d5\u5dee(Sun et al., 2007)\uff1a \u5f37\u5065\u6027\u6548\u679c\u3002NMF \u7684\u6578\u5b78\u5f0f\u8868\u793a\u5982\u4e0b\uff1a</td></tr><tr><td/><td>|</td><td>|</td><td>(4)</td></tr><tr><td colspan=\"4\">\u5728\u5f0f(4)\u4e2d\uff0c \u2027 \u70ba\u55ae\u4e00\u8a9e\u53e5\u67d0\u4e00\u7279\u5fb5\u7dad\u5ea6\u7684\u8abf\u8b8a\u983b\u8b5c\u5f37\u5ea6\u4e4b\u7d2f\u7a4d\u5206\u5e03\u51fd\u6578(Cumulative</td></tr><tr><td colspan=\"2\">Distribution Function, CDF)\uff0c</td><td colspan=\"2\">\u3002\u7576\u8a9e\u97f3\u8a0a \u5247\u662f\u5229\u7528\u6240\u6709\u8a13\u7df4\u8a9e\u53e5\u4e4b\u8abf\u8b8a\u983b\u8b5c\u5f37\u5ea6\u6240\u6c42\u5f97\u7684\u5c0d\u61c9\u4e4b</td></tr><tr><td colspan=\"4\">\u865f\u53d7\u5230\u96dc\u8a0a\u5f71\u97ff\u6642\uff0c\u5176\u8a9e\u97f3\u7279\u5fb5\u6642\u9593\u5e8f\u5217\u6703\u53d7\u5230\u5f71\u97ff\u800c\u5931\u771f\uff0c\u53ca\u5176\u8abf\u8b8a\u983b\u8b5c\u4e5f\u6703\u8ddf\u8457\u53d7 \u53c3\u8003\u7d2f\u7a4d\u5206\u5e03\u51fd\u6578\uff0c \u4fbf\u662f\u66f4\u65b0\u904e\u5f8c\u7684\u8abf\u8b8a\u983b\u8b5c\u5f37\u5ea6\u6210\u5206\u3002</td></tr><tr><td colspan=\"4\">\u5230\u727d\u9023\u3002\u5f88\u591a\u5b78\u8005\u63d0\u51fa\u4f5c\u7528\u5728\u8abf\u8b8a\u983b\u8b5c\u7684\u6b63\u898f\u5316\u6cd5\uff0c\u4ee5\u6539\u5584\u8abf\u8b8a\u983b\u8b5c\u53d7\u5230\u96dc\u8a0a\u5e72\u64fe\u7684\u5f71</td></tr><tr><td colspan=\"4\">\u97ff\u3002\u56e0\u6b64\uff0c\u6211\u5011\u53ef\u5c07\u8a31\u591a\u767c\u5c55\u5728\u8a9e\u97f3\u7279\u5fb5\u6642\u9593\u5e8f\u5217\u7684\u6b63\u898f\u5316\u6cd5\u61c9\u7528\u5728\u8abf\u8b8a\u983b\u8b5c\u4f7f\u5176\u6b63\u898f 2.5 \u5206\u983b\u6bb5\u8abf\u8b8a\u983b\u8b5c\u7d71\u8a08\u6b63\u898f\u5316\u6cd5</td></tr><tr><td colspan=\"4\">\u5316\uff1b\u800c\u6b63\u898f\u5316\u7684\u5c0d\u8c61\u662f\u5c0d\u5176\u8abf\u8b8a\u983b\u8b5c\u5f37\u5ea6(Magnitude)\u6210\u5206|X[k]|\u4f86\u9032\u884c\u8655\u7406\uff0c\u4e26\u4fdd\u6301\u5176\u76f8 \u6b64\u65b9\u6cd5\u7684\u6982\u5ff5\u662f\u60f3\u8981\u6539\u9032\u539f\u59cb\u8abf\u8b8a\u983b\u8b5c\u7d71\u8a08\u6b63\u898f\u5316\u6cd5\uff1b\u539f\u59cb\u8abf\u8b8a\u983b\u8b5c\u7d71\u8a08\u6b63\u898f\u5316\u6cd5\u662f\u5c07</td></tr><tr><td colspan=\"4\">\u4f4d\u89d2\u4e0d\u8b8a\u03b8[k]=\u2220X[k]\u7684\u90e8\u5206\u3002\u63a5\u8457\uff0c\u7d93\u8655\u7406\u5f8c\u88ab\u66f4\u65b0\u7684\u5f37\u5ea6\u6210\u5206\u6703\u8207\u539f\u59cb\u76f8\u4f4d\u6210\u5206\u7d50\u5408\uff0c \u5168\u90e8\u8abf\u8b8a\u983b\u5e36\u7684\u983b\u8b5c\u5f37\u5ea6\u503c\u8996\u70ba\u662f\u5c6c\u65bc\u540c\u4e00\u96a8\u6a5f\u8b8a\u6578\u7684\u6a23\u672c(Samples)\uff0c\u4e14\u5c07\u4e4b\u4e00\u4f75\u9032\u884c</td></tr><tr><td colspan=\"4\">\u518d\u85c9\u7531\u53cd\u5085\u7acb\u8449\u8f49\u63db(Inverse Discrete Fourier Transform, IDFT)\u4f86\u6c42\u5f97\u65b0\u7684\u8a9e\u97f3\u7279\u5fb5\u6642\u9593 \u6b63\u898f\u5316\u7684\u52d5\u4f5c\u3002\u4f46\u662f\u524d\u9762\u63d0\u5230\u5728\u8a9e\u97f3\u8fa8\u8b58\u4e2d\uff0c\u4e0d\u540c\u8abf\u8b8a\u983b\u7387\u7684\u6210\u5206\u6709\u4e0d\u540c\u7684\u91cd\u8981\u6027\uff0c\u4f4e</td></tr><tr><td colspan=\"4\">\u5e8f\u5217\u3002\u82e5\u8abf\u8b8a\u983b\u8b5c\u7684\u5f37\u5ea6\u80fd\u5920\u88ab\u6709\u6548\u7684\u6b63\u898f\u5316\uff0c\u4fbf\u80fd\u5920\u6709\u6548\u89e3\u6c7a\u96dc\u8a0a\u7522\u751f\u7684\u74b0\u5883\u4e0d\u5339\u914d \u983b\u6210\u5206\u662f\u6bd4\u9ad8\u983b\u6210\u5206\u9084\u8981\u76f8\u5c0d\u91cd\u8981\u7684\uff0c\u56e0\u70ba\u8a9e\u8a00\u7684\u91cd\u8981\u8cc7\u8a0a\u8f03\u96c6\u4e2d\u65bc\u4f4e\u983b\u6210\u5206\u3002\u56e0\u6b64\uff0c</td></tr><tr><td colspan=\"4\">\u554f\u984c\uff0c\u4f7f\u81ea\u52d5\u8a9e\u97f3\u8fa8\u8b58\u7cfb\u7d71\u5728\u4f7f\u7528\u65b0\u7684\u8a9e\u97f3\u7279\u5fb5\u7684\u60c5\u6cc1\u4e0b\u80fd\u5920\u7372\u5f97\u8f03\u4f73\u7684\u8fa8\u8b58\u7387\u3002\u4ee5\u4e0b \u6709\u5b78\u8005\u63d0\u51fa\u5c07\u8abf\u8b8a\u983b\u5e36\u5206\u6210\u8a31\u591a\u5b50\u983b\u6bb5\uff0c\u518d\u5206\u5225\u5c0d\u6bcf\u4e00\u500b\u5b50\u983b\u6bb5\u7684\u983b\u8b5c\u5f37\u5ea6\u4f5c\u4e0a\u8ff0\u6240\u63d0</td></tr><tr><td colspan=\"4\">\u5c07\u6703\u7c21\u55ae\u56de\u9867\u4e00\u4e9b\u5e38\u898b\u7684\u8abf\u8b8a\u983b\u8b5c\u6b63\u898f\u5316\u6cd5\u3002 \u7684\u8abf\u8b8a\u983b\u8b5c\u6b63\u898f\u5316\u7684\u65b9\u6cd5\uff0c\u800c\u4e0d\u662f\u55ae\u7d14\u76f4\u63a5\u5c0d\u6574\u500b\u5168\u90e8\u8abf\u8b8a\u983b\u5e36\u505a\u8655\u7406(Huang et al.,</td></tr><tr><td colspan=\"4\">2009)\u3002\u56e0\u70ba\u8981\u5f37\u8abf\u4f4e\u8abf\u8b8a\u983b\u7387\u7684\u91cd\u8981\u6027\uff0c\u6240\u4ee5\u5728\u4f4e\u983b\u90e8\u5206\u7684\u5b50\u983b\u6bb5\u64c1\u6709\u8f03\u7a84\u7684\u983b\u5bec\uff0c\u5b50</td></tr><tr><td colspan=\"4\">\u983b\u6bb5\u7684\u6578\u91cf\u4e5f\u6bd4\u8f03\u591a\uff0c\u800c\u9ad8\u8abf\u8b8a\u983b\u7387\u4fbf\u6301\u6709\u76f8\u53cd\u7684\u7279\u6027\u3002\u7531\u65bc\u80fd\u66f4\u7d30\u7dfb\u5730\u5206\u6790\u8207\u8655\u7406\u4f4e</td></tr><tr><td colspan=\"4\">\u5047\u8a2d\u7576\u5404\u7a2e\u97f3\u7d20\u5728\u4e00\u822c\u74b0\u5883\u4e2d\u5206\u5e03\u7684\u6bd4\u4f8b\u63a5\u8fd1\u4e00\u81f4\u6642\uff0c\u6bcf\u4e00\u7dad\u5ea6\u8a9e\u97f3\u7279\u5fb5\u7684\u8abf\u8b8a\u983b\u8b5c\u4e4b \u983b\u6210\u5206\u7684\u8cc7\u8a0a\uff0c\u904e\u53bb\u7684\u4e00\u4e9b\u5be6\u9a57\u6578\u64da\u986f\u793a\u51fa\u5c07\u8abf\u8b8a\u983b\u7387\u5206\u983b\u6bb5\u4f86\u6b63\u898f\u5316\u7684\u505a\u6cd5\uff0c\u80fd\u6bd4\u5168</td></tr><tr><td colspan=\"3\">\u5e73\u5747\u503c\u61c9\u8a72\u70ba\u4e00\u500b\u5b9a\u503c(Huang et al., 2009)\uff1a \u983b\u5e36\u6b63\u898f\u5316\u7684\u65b9\u5f0f\u7372\u5f97\u8f03\u597d\u7684\u6548\u80fd\u3002</td></tr><tr><td colspan=\"3\">| 3. \u4e09\u7a2e\u65b0\u7a4eNMF\u6539\u9032\u65b9\u6cd5\u7528\u65bc\u8abf\u8b8a\u983b\u8b5c\u5206\u89e3 |</td><td>(2)</td></tr><tr><td>\u5728\u5f0f(2)\u4e2d\uff0c|</td><td colspan=\"3\">|\u70ba\u539f\u59cb\u7684\u8abf\u8b8a\u983b\u8b5c\u5f37\u5ea6\u6210\u5206\uff0c \u70ba\u55ae\u4e00\u8a9e\u53e5\u7684\u8abf\u8b8a\u983b\u8b5c\u5f37\u5ea6\u6210\u5206\u4e4b\u5e73</td></tr><tr><td colspan=\"3\">\u5747\u503c\uff0c \u70ba\u6240\u6709\u8a13\u7df4\u8a9e\u53e5\u7684\u8abf\u8b8a\u983b\u8b5c\u5f37\u5ea6\u6210\u5206\u4e4b\u5e73\u5747\u503c\uff0c\u800c 3.1 \u50b3\u7d71\u975e\u8ca0\u77e9\u9663\u5206\u89e3\u6cd5(NMF)</td><td>\u4fbf\u662f\u66f4\u65b0\u904e\u5f8c\u7684\u8abf\u8b8a\u983b</td></tr><tr><td colspan=\"4\">\u8b5c\u5f37\u5ea6\u6210\u5206\u3002 \u5728\u5f88\u591a\u9818\u57df\u4e2d\u5982\u4f55\u5c0b\u627e\u91cd\u8981\u7684\u6f5b\u85cf\u8cc7\u8a0a\u6210\u5206\u662f\u500b\u91cd\u8981\u7684\u8b70\u984c\uff0c\u800c\u57fa\u65bc\u975e\u8ca0\u77e9\u9663\u5206\u89e3\u6cd5</td></tr><tr><td colspan=\"4\">(Nonnegative Matrix Factorization, NMF)(Lee & Seung, 1999)\u7684\u6280\u8853\u53ef\u4ee5\u88ab\u7528\u65bc\u8655\u7406\u6b64\u8b70</td></tr><tr><td colspan=\"4\">\u984c\u3002\u9867\u540d\u601d\u7fa9\uff0c\u6b64\u65b9\u6cd5\u5c31\u662f\u5c07\u975e\u8ca0\u7684\u539f\u59cb\u8cc7\u6599\u6240\u6210\u7684\u77e9\u9663\u9032\u884c\u5206\u89e3\uff0c\u8868\u793a\u6210\u5169\u500b\u4e5f\u662f\u975e \u8ca0\u7684\u77e9\u9663\u4e58\u7a4d\uff0c\u63a5\u8457\u5229\u7528\u7dda\u6027\u7d44\u5408\u7684\u7279\u6027\u4f86\u8868\u793a\u539f\u59cb\u8cc7\u6599\u4e2d\u5404\u500b\u6a23\u672c\u4e4b\u76ee\u7684\u3002\u800c\u5176\u5b83\u5e38 2.3 \u8abf \u8b8a \u983b \u8b5c \u5e73 \u5747 \u8207 \u8b8a \u7570 \u6578 \u6b63 \u898f \u5316 \u6cd5 \u9664\u4e86\u8981\u6b63\u898f\u5316\u8abf\u8b8a\u983b\u8b5c\u5f37\u5ea6\u6210\u5206\u4e4b\u5e73\u5747\u503c\u5916\uff0c\u4e5f\u53ef\u540c\u6642\u6b63\u898f\u5316\u5176\u6a19\u6e96\u5dee(Huang et al., \u898b\u7684\u7dda\u6027\u8868\u793a\u6cd5\u6709\u4e3b\u6210\u5206\u5206\u6790(Principal Component Analysis, PCA)\u8207\u7368\u7acb\u6210\u5206\u5206\u6790</td></tr><tr><td colspan=\"4\">2009)\u3002\u5047\u8a2d\u7279\u5fb5\u5411\u91cf\u53c3\u6578\u4e4b\u5e73\u5747\u503c\u8207\u8b8a\u7570\u6578\u5728\u4e00\u822c\u74b0\u5883\u4e2d\u5206\u5e03\u7684\u6bd4\u4f8b\u63a5\u8fd1\u4e00\u81f4\u6642\uff0c\u6211\u5011 (Independent Component Analysis, ICA)\u3002\u975e\u8ca0\u77e9\u9663\u5206\u89e3\u6cd5\u8207\u9019\u5169\u7a2e\u7dda\u6027\u8868\u793a\u6cd5\u4e4b\u5dee\u7570\u5c31\u662f</td></tr></table>", | |
| "num": null | |
| }, | |
| "TABREF5": { | |
| "html": null, | |
| "text": "HEQ \u7686\u662f\u5728\u8a9e\u53e5\u7684\u97f3\u6846\u5c64\u9762(Frame Level)\u5c0d\u6bcf\u500b\u97f3\u6846 \u5206\u5225\u4f5c\u6b63\u898f\u5316\uff0c\u800c NMF \u7684\u65b9\u6cd5\u662f\u5728\u6574\u9ad4\u8a9e\u53e5\u5c64\u6b21(Utterance Level)\u6b63\u898f\u5316\uff0c\u56e0\u5206\u5225\u8655\u7406 \u4e0d\u540c\u7684\u9762\u5411\uff0c\u6240\u4ee5\u5728\u7d50\u5408\u5f8c\u6709\u52a0\u6210\u6027\u7684\u6548\u679c\u3002\u6548\u679c\u63d0\u5347\u6700\u986f\u8457\u7684\u662f CMVN \u8207 NSHGNMF \u7684\u7d50\u5408\uff1b\u5176\u6b21\u662f\u8207 HEQ \u7d50\u5408\u7684 NSHGNMF\uff0c\u7686\u80fd\u6709\u4e0d\u932f\u7684\u9032\u6b65\u3002\u6211\u5011\u4e5f\u5c07\u6240\u63d0\u51fa\u65b9\u6cd5 \u8207\u8207\u9032\u968e\u524d\u7aef\u6a19\u6e96(Advanced Front-End Standard, AFE)\u8655\u7406\u904e\u8a9e\u97f3\u7279\u5fb5(Macho et al.,", | |
| "type_str": "table", | |
| "content": "<table><tr><td>96 98</td><td>\u8abf\u8b8a\u983b\u8b5c\u5206\u89e3\u6280\u8853\u65bc\u5f37\u5065\u8a9e\u97f3\u8fa8\u8b58\u4e4b\u7814\u7a76 \u8abf\u8b8a\u983b\u8b5c\u5206\u89e3\u6280\u8853\u65bc\u5f37\u5065\u8a9e\u97f3\u8fa8\u8b58\u4e4b\u7814\u7a76 \u8abf\u8b8a\u983b\u8b5c\u5206\u89e3\u6280\u8853\u65bc\u5f37\u5065\u8a9e\u97f3\u8fa8\u8b58\u4e4b\u7814\u7a76</td><td>\u5f35\u5ead\u8c6a \u7b49 97 \u5f35\u5ead\u8c6a \u7b49 \u5f35\u5ead\u8c6a \u7b49 101 \u5f35\u5ead\u8c6a \u7b49 103</td></tr><tr><td colspan=\"3\">\u76ee\u7684\u3002\u5c07\u4e0a\u8ff0\u6240\u6c42\u51fa\u7684 \u7576\u4f5c\u61f2\u7f70\u9805\uff0c\u52a0\u5165\u5230\u50b3\u7d71 NMF \u4e4b\u6b50\u5f0f\u8ddd\u96e2\u6e1b\u640d\u51fd\u5f0f\u4e2d\u53ef\u4ee5\u5f97 4.3 \u8fa8\u8b58\u6548\u80fd\u8a55\u4f30\u65b9\u5f0f 4.5 \u57fa\u65bc\u5716\u6b63\u5247\u5316\u975e\u8ca0\u77e9\u9663\u5206\u89e3\u6cd5(GNMF)\u4e4b\u5be6\u9a57\u7d50\u679c \u6578\u64da\u986f\u793a\u8a9e\u97f3\u8fa8\u8b58\u7684\u8a5e\u6703\u96a8 \u503c\u8b8a\u5927\u800c\u9010\u6f38\u63d0\u9ad8\u3002\u53ef\u80fd\u662f\u56e0\u70ba\u82e5 \u7684\u503c\u8d8a\u4f4e\uff0c\u9580\u6abb\u8d8a\u56b4\u683c\uff0c 4.6 \u975e\u8ca0\u7de8\u78bc\u77e9\u9663\u7d71\u8a08\u5716\u7b49\u5316\u6cd5(HNMF)\u4e4b\u5be6\u9a57\u7d50\u679c HEQ \u5c07\u8abf\u8b8a\u983b\u8b5c\u5f37\u5ea6\u6210\u5206\u7684\u5e73\u5747\u503c\u8207\u6a19\u6e96\u5dee\u6b63\u898f\u5316\uff0c\u4e26\u540c\u6642\u6b63\u898f\u5316\u5176\u5b83\u968e\u5c64\u7684\u52d5\u5dee\u4f7f\u8a13 4.8 \u7d50\u5408\u4e0d\u540c\u6642\u9593\u5e8f\u5217\u6b63\u898f\u5316\u6cd5\u4e4b\u7d50\u679c</td></tr><tr><td colspan=\"3\">\u5230\u65b0\u7684\u6e1b\u640d\u51fd\u6578\uff1a \u8fa8\u8b58\u6548\u80fd\u7684\u8a55\u4f30\u65b9\u5f0f\u662f\u4f9d\u7167\u7f8e\u570b\u570b\u5bb6\u6a19\u6e96\u8207\u79d1\u6280\u5c40(National Institute of Standards and \u5728\u63a2\u8a0e GNMF \u7684\u6548\u80fd\u6642\uff0c\u6211\u5011\u9996\u5148\u5728\u6c42\u53d6\u6b0a\u91cd\u77e9\u9663 E \u6642\u4f7f\u7528 0-1 \u6b0a\u91cd\u7684\u65b9\u5f0f\uff1b\u70ba\u6b64\uff0c\u6211 \u6b0a\u91cd\u77e9\u9663E\u4e2d\u975e\u96f6\u503c\u7684\u5143\u7d20\u5c31\u8d8a\u5c11\uff0c\u5c0e\u81f4\u4efb\u5169\u8a9e\u53e5\u9593\u7684\u95dc\u806f\u5ea6\u5728\u6c42\u53d6\u77e9\u9663W \u8207 H\u6642\u8f03\u4e0d \u5982\u7b2c\u4e09\u7bc0\u6240\u63d0\u53ca\uff0c\u6211\u5011\u9032\u4e00\u6b65\u4fdd\u7559\u5728\u8a13\u7df4\u968e\u6bb5\u7372\u5f97\u4e4b\u4e7e\u6de8\u8a13\u7df4\u8a9e\u6599\u7de8\u78bc\u77e9\u9663 H \u7684\u7d2f\u7a4d\u5206 \u7df4\u8a9e\u53e5\u8207\u6e2c\u8a66\u8a9e\u53e5\u7684\u8abf\u8b8a\u983b\u8b5c\u5f37\u5ea6\u7684\u6a5f\u7387\u5206\u5e03\u8da8\u65bc\u4e00\u81f4\u3002PCA \u5247\u662f\u5c0d\u6240\u6709\u8a13\u7df4\u8a9e\u53e5\u7684\u8abf \u6700\u5f8c\u6211\u5011\u63a2\u8a0e\u984d\u5916\u7d50\u5408\u4e0d\u540c\u6642\u9593\u5e8f\u5217\u6b63\u898f\u5316\u6cd5(CMVN \u8207 HEQ)\u8207\u672c\u8ad6\u6587\u6240\u63d0\u51fa\u4e09\u7a2e\u8abf\u8b8a</td></tr><tr><td colspan=\"3\">\u2016V WH\u2016 Technology, NIST)\u6240\u8a02\u7acb\u7684\u8a55\u4f30\u6a19\u6e96\uff0c\u9032\u884c\u6bcf\u4e00\u53e5\u6e2c\u8a66\u8a9e\u53e5\u4e4b\u6b63\u78ba\u8f49\u5beb\u8a5e\u4e32\u8207\u8a9e\u97f3\u8fa8\u8b58 HLH (18) \u5011\u5148\u7b97\u51fa\u6240\u6709\u8a13\u7df4\u8a9e\u53e5(8,440 \u53e5)\u5f7c\u6b64\u9593\u7684\u95dc\u806f\u5ea6\u3002\u95dc\u806f\u5ea6\u7684\u4f30\u6e2c\u662f\u900f\u904e\u8a08\u7b97\u5169\u5169\u8a13\u7df4\u8a9e \u6703\u88ab\u5f37\u8abf\u3002 \u5e03\u51fd\u6578(CDF)\u8cc7\u8a0a\uff0c\u5c07\u5176\u5132\u5b58\u5efa\u8868\u4ee5\u4f9b\u6e2c\u8a66\u968e\u6bb5\u4f7f\u7528\u7d71\u8a08\u5716\u7b49\u5316\u6cd5\u4f86\u6b63\u898f\u5316\u6bcf\u4e00\u6e2c\u8a66\u8a9e \u8b8a\u983b\u8b5c\u5f37\u5ea6\u6210\u5206\u6c42\u53d6\u5171\u8b8a\u7570\u6578\uff0c\u63a5\u8457\u5229\u7528\u524d r \u500b\u7279\u5fb5\u503c(Eigenvalues)\u53bb\u627e\u5176\u5c0d\u61c9\u7684 r \u500b\u7279 \u983b\u8b5c\u975e\u8ca0\u77e9\u9663\u5206\u89e3\u6cd5\u7684\u5be6\u9a57\u7d50\u679c\uff0c\u5982\u8868 7 \u8207 8 \u6240\u793a\u3002\u672c\u8ad6\u6587\u6240\u63d0\u51fa\u4e09\u7a2e\u8abf\u8b8a\u983b\u8b5c\u975e\u8ca0\u77e9</td></tr><tr><td colspan=\"3\">\u8a5e\u4e32\u7684\u6bd4\u8f03\u3002\u8a55\u4f30\u65b9\u5f0f\u662f\u4ee5\u8a5e\u6b63\u78ba\u7387(Word Accuracy Rate)\u70ba\u4e3b\uff0c\u8a08\u7b97\u6b63\u78ba\u8f49\u5beb\u8a5e\u4e32\u8207\u8a9e \u53e5\u5f7c\u6b64\u9593\u7684\u97f3\u7d20\u932f\u8aa4\u7387(Phone Error Rate, PER)\u800c\u5f97\uff1b\u6211\u5011\u6703\u5148\u6c42\u5f97\u6bcf\u4e00\u53e5\u8a13\u7df4\u8a9e\u53e5\u7d93\u4eba \u8868 3. GNMF-a \u4f7f\u7528\u6b0a\u91cd\u77e9\u9663\u5168\u57df\u7d66\u503c\u4e4b\u8a5e\u6b63\u78ba\u7387(%) \u53e5\u7684\u7de8\u78bc\u5411\u91cf\u3002\u5176\u6578\u64da\u5982\u8868 5 \u6240\u793a\uff0c\u6b64\u65b9\u6cd5(HNMF)\u5728\u57fa\u5e95\u500b\u6578\u7b49\u65bc 5 \u6642\uff0c\u53ef\u4ee5\u9054\u5230\u512a\u65bc \u5fb5\u5411\u91cf(Eigenvectors)\u4ee5\u7576\u4f5c\u8abf\u8b8a\u983b\u8b5c\u5f37\u5ea6\u6210\u5206\u7684 PCA \u5b50\u7a7a\u9593\u4e4b\u57fa\u5e95\uff0c\u4f7f\u6e2c\u8a66\u8a9e\u53e5\u7684\u8abf\u8b8a \u9663\u5206\u89e3\u6cd5\u90fd\u80fd\u8207\u5148\u7d93\u904e\u4e0d\u540c\u6642\u9593\u5e8f\u5217\u6b63\u898f\u5316\u6cd5\u8655\u7406\u904e\u5f8c\u7684\u8a9e\u97f3\u7279\u5fb5\u76f8\u7d50\u5408\u4f7f\u7528\u800c\u5f97\u5230\u6548 \u540c\u6a23\u5730\uff0c\u53ef\u5229\u7528\u68af\u5ea6\u4e0b\u964d\u6f14\u7b97\u6cd5\u53bb\u6c42\u51fa\u57fa\u65bc\u5716\u6b63\u5247\u5316\u975e\u8ca0\u77e9\u9663\u5206\u89e3\u6cd5\u7684\u4e58\u6cd5\u66f4\u65b0\u898f\u5247\uff1a H \u2190 H W V HE W WH HD W \u2190 W VH WHH (19) \u5176\u4e2d 0\uff0c\u70ba\u6b63\u5247\u5316\u53c3\u6578\uff0c\u53bb\u63a7\u5236\u65b0\u7684\u8868\u793a\u4e4b\u5e73\u6ed1\u6027\u3002 \u6709\u5225\u65bc\u50b3\u7d71 NMF \u65b9\u6cd5\u50c5\u5728\u6b50\u6c0f\u7a7a\u9593\u4e2d\u6c42\u89e3\uff0cGNMF \u65b9\u6cd5\u53ef\u4ee5\u8996\u4e0d\u540c\u61c9\u7528\u554f\u984c\u800c\u8a2d \u8a08\u5408\u9069\u7684\u6b0a\u91cd\u77e9\u9663 E\u3002\u5728\u8a9e\u97f3\u8fa8\u8b58\u7684\u4efb\u52d9\u4e2d\uff0c\u8072\u5b78\u6a21\u578b\u901a\u5e38\u88ab\u5efa\u7acb\u5728\u97f3\u7d20\u5c64\u6b21\uff0c\u800c\u4e14\u4e3b \u5bb0\u8a9e\u97f3\u8fa8\u8b58\u7684\u8868\u73fe\u3002\u56e0\u6b64\uff0c\u672c\u8ad6\u6587\u4e5f\u63d0\u51fa\u5229\u7528\u97f3\u7d20\u932f\u8aa4\u7387\u5efa\u7acb\u6b0a\u91cd\u77e9\u9663 E\uff0c\u8a73\u7d30\u7684\u63cf\u8ff0 \u8207\u5be6\u9a57\u7a0d\u5f8c\u5c07\u88ab\u5448\u73fe\u5728\u7b2c 4.3 \u7bc0\u3002 \u5716 4. \u975e\u8ca0\u7de8\u78bc\u77e9\u9663\u7d71\u8a08\u5716\u7b49\u5316\u6cd5(HNMF)\u9084\u539f\u793a\u610f\u5716 4. \u5be6\u9a57\u7d50\u679c\u8207\u5206\u6790 4.1 \u5be6\u9a57\u8a9e\u6599\u5eab \u672c \u8ad6 \u6587 \u5be6 \u9a57 \u6240 \u63a1 \u7528 \u7684 \u8a9e \u6599 \u5eab \u662f Aurora-2 \uff0c \u5b83 \u662f \u7531 \u6b50 \u6d32 \u96fb \u4fe1 \u6a19 \u6e96 \u5354 \u6703 (European Telecommunications Standards Institute, ETSI)\u6240\u767c\u884c\u7684\u8a9e\u6599\u5eab(Hirsch & Pearce, 2000)\uff0c\u4ee5 \u7f8e\u570b\u6210\u5e74\u4eba\u7684\u8072\u97f3\u4f5c\u70ba\u9304\u97f3\u4f86\u6e90\uff0c\u5167\u5bb9\u662f\u9023\u7e8c\u7684\u82f1\u6587\u6578\u5b57\u7531 0(Zero)\u5230 9(Nine)\u8ddf Oh \u7b49\u767c \u97f3\u5b57\u8a5e\u3002\u8a9e\u6599\u5eab\u5167\u6709\u4e7e\u6de8\u53ca\u542b\u6709\u96dc\u8a0a\u7684\u8a9e\u97f3\uff0c\u96dc\u8a0a\u4e2d\u6709\u516b\u7a2e\u4e0d\u540c\u7684\u52a0\u6210\u6027\u96dc\u8a0a\u8207\u5169\u7a2e\u4e0d \u540c\u7684\u901a\u9053\u6548\u61c9\uff0c\u800c\u901a\u9053\u6548\u61c9\u662f\u4f7f\u7528\u570b\u969b\u96fb\u4fe1\u806f\u5408\u6703(ITU)\u6a19\u6e96\u4e2d\u7684 G.712 \u548c MIRS\u3002\u6839\u64da \u97f3\u8fa8\u8b58\u8a5e\u4e32\u5f7c\u6b64\u9593\u7684\u8a5e\u53d6\u4ee3\u500b\u6578(Substitutions)\u3001\u8a5e\u63d2\u5165\u500b\u6578(Insertions)\u548c\u8a5e\u522a\u9664\u500b\u6578 (Deletions)\uff1a \u8a5e\u6b63\u78ba\u7387 % \u8a5e\u6b63\u78ba\u8fa8\u8b58\u500b\u6578 \u8a5e\u63d2\u5165\u500b\u6578 \u8f38\u5165\u8a5e\u7e3d\u6578 100% (20) \u6700\u5f8c\u5728\u8a55\u4f30\u6574\u9ad4\u8a9e\u97f3\u8fa8\u8b58\u6548\u80fd\u6642\uff0c\u6211\u5011\u53c3\u7167\u570b\u969b\u5b78\u8005\u4e4b\u8a2d\u5b9a\uff0c\u5c0d\u6e2c\u8a66\u8a9e\u53e5\u5728\u6bcf\u4e00\u7a2e\u566a\u97f3 \u7684\u8a0a\u566a\u6bd4\u7684\u8a5e\u6b63\u78ba\u7387\u7d50\u679c\u505a\u52a0\u7e3d\u8207\u53d6\u5e73\u5747\u7684\u52d5\u4f5c(\u53bb\u6389\u6975\u7aef\u7684\u8a0a\u566a\u6bd4 Clean \u8ddf-5\uff0c\u53ea\u8a08\u7b97 \u7bc4\u570d 20dB \u5230 0dB \u4e2d\u7684\u5e73\u5747\u8a5e\u6b63\u78ba\u7387)\uff1b\u672c\u8ad6\u6587\u4ee5\u4e0b\u7684\u5168\u90e8\u5be6\u9a57\u7686\u662f\u5229\u7528\u5e73\u5747\u8a5e\u6b63\u78ba\u7387\u4f86 \u8a55\u4f30\u8a9e\u97f3\u8fa8\u8b58\u7684\u6548\u80fd\u3002 4.4 \u975e\u5e73\u6ed1\u975e\u8ca0\u77e9\u9663\u5206\u89e3\u6cd5(NSNMF)\u4e4b\u5be6\u9a57\u7d50\u679c \u5de5\u8f49\u5beb(Transcription)\u4e4b\u97f3\u7d20\u5e8f\u5217(Phone Sequence)\u3002\u672c\u8ad6\u6587\u97f3\u7d20\u932f\u8aa4\u7387\u7684\u7b97\u6cd5\u662f\u4f7f\u7528\u7de8\u8f2f \u8ddd\u96e2(Edit Distance)\u7684\u7b97\u6cd5\uff0c\u8a08\u7b97\u6bcf\u4e00\u53e5\u8a13\u7df4\u8a9e\u53e5(\u7576\u6210\u76ee\u6a19\u8a9e\u53e5)\u7684\u97f3\u7d20\u5e8f\u5217\u8207\u5176\u5b83\u8a13\u7df4 \u8a9e\u53e5\u7684\u97f3\u7d20\u5e8f\u5217\u5f7c\u6b64\u9593\u7684\u97f3\u7d20\u53d6\u4ee3\u500b\u6578\u3001\u97f3\u7d20\u63d2\u5165\u500b\u6578\u3001\u97f3\u7d20\u522a\u9664\u500b\u6578\uff0c\u4e26\u4f9d\u4e0b\u5f0f\u8a08\u7b97 \u97f3\u7d20\u932f\u8aa4\u7387\uff1a \u97f3\u7d20\u932f\u8aa4\u7387 % \u97f3\u7d20\u53d6\u4ee3\u6578 \u97f3\u7d20\u63d2\u5165\u6578 \u97f3\u7d20\u522a\u9664\u6578 \u76ee\u6a19\u8a13\u7df4\u8a9e\u53e5\u97f3\u7d20\u7e3d\u6578 100% (21) \u6240\u4ee5\u6700\u5f8c\u6c42\u5f97\u6b0a\u91cd\u77e9\u9663 E \u662f\u500b\u7dad\u5ea6\u5927\u5c0f\u70ba 8,400*8,400 \u7684\u77e9\u9663\uff0c\u5176\u4e2d\u6bcf\u500b\u5143\u7d20\u90fd\u7d00\u9304\u8457 \u6bcf\u4e00\u53e5\u8a13\u7df4\u8a9e\u53e5\u8207\u5176\u5b83\u8a9e\u53e5\u5f7c\u6b64\u9593\u7684\u97f3\u7d20\u932f\u8aa4\u7387\uff0c\u5c0d\u89d2\u7dda\u4e0a\u7684\u70ba\u4e00\u500b\u4f4d\u7f6e\u70ba\u67d0\u4e00\u53e5\u8a13\u7df4 \u8a9e\u53e5\u81ea\u5df1\u672c\u8eab\u6240\u4ee5\u5dee\u7570\u662f 0\u3002\u503c\u5f97\u4e00\u63d0\u7684\u662f\uff0c\u7576\u4f7f\u7528\u7de8\u8f2f\u8ddd\u96e2\u7b97\u5dee\u7570\u5ea6\u6642\uff0cE \u8207E \u7684\u503c \u53ef\u80fd\u4e0d\u6703\u4e00\u6a23\uff0c\u662f\u56e0\u70ba\u4e0d\u540c\u7684\u76ee\u6a19\u8a9e\u53e5(\u4efb\u5169\u53e5\u8a13\u7df4\u8a9e\u53e5\uff0c\u5f7c\u6b64\u7684\u97f3\u7d20\u5e8f\u5217\u9577\u5ea6\u53ef\u80fd\u6703\u662f \u4e0d\u540c\u7684)\u7684\u5047\u5b9a\uff0c\u6240\u4ee5\u6b0a\u91cd\u77e9\u9663 E \u662f\u4e0d\u5c0d\u7a31\u7684\u3002\u4f46\u6211\u5011\u8a8d\u70ba\u5169\u500b\u8a13\u7df4\u8a9e\u53e5\u9593\u5f7c\u6b64\u7684\u95dc\u806f\u5ea6 Set A Set B Set C Average NMF 67.09 70.98 68.22 68.87 \u03b1 0.9 68.49 72.64 68.00 70.05 GNMF-a 70.63 74.27 70.78 72.12 \u975e\u5e73\u6ed1\u975e\u8ca0\u77e9\u9663\u5206\u89e3\u6cd5(NSNMF)\u7684\u6548\u80fd\u3002\u4f46\u53ef\u767c\u73fe\u82e5\u57fa\u5e95\u500b\u6578\u82e5\u6301\u7e8c\u589e\u52a0\u6642\uff0c\u4f3c\u4e4e\u5c31\u7121 \u6cd5\u8207 NSNMF \u7af6\u722d\uff0c\u4e14\u6548\u80fd\u63d0\u6607\u7684\u7a0b\u5ea6\u4e0d\u660e\u986f\u3002 \u8868 5. HNMF \u4e4b\u4e0d\u540c\u57fa\u5e95\u500b\u6578\u4e4b\u8a5e\u6b63\u78ba\u7387(%) Set A Set B Set C Average K=5 77.65 80.16 77.25 78.57 \u983b\u8b5c\u5f37\u5ea6\u6210\u5206\u80fd\u5920\u6295\u5f71\u5230 PCA \u5b50\u7a7a\u9593\u4ee5\u9054\u5230\u6b63\u898f\u5316\u7684\u76ee\u7684\u3002 \u8868 7. \u7d50\u5408 CMVN \u8207 NMF \u4e4b\u8a5e\u6b63\u78ba\u7387(%) Set A Set B Set C Average CMVN 75.93 76.76 76.82 76.44 CMVN+NSNMF 83.56 85.51 83.27 84.28 \u80fd\u63d0\u6607\u3002\u503c\u5f97\u6ce8\u610f\u7684\u662f\uff0cCMVN \u8207 2002)\u505a\u7d50\u5408\uff0c\u5176\u7d50\u679c\u5982\u8868 9 \u6240\u793a\u3002AFE \u662f\u8fd1\u5e74\u4f86\u6b50\u6d32\u96fb\u4fe1\u6a19\u6e96\u5354\u6703(ETSI)\u6240\u63a8\u51fa\u7684\u7279\u5fb5 \u57fa\u65bc\u6b64\u89c0\u5bdf\uff0c\u672c\u8ad6\u6587\u5617\u8a66\u6539\u6210\u8b93\u6b0a\u91cd\u77e9\u9663E\u7684\u6bcf\u4e00\u500b\u5143\u7d20\u90fd\u80fd\u64c1\u6709\u9069\u7576\u7684\u6b0a\u91cd\u503c\uff0c\u800c\u4e0d\u4f7f K=10 69.55 74.32 67.77 71.10 CMVN+GNMF 83.58 84.78 82.36 83.81 \u5411\u91cf\u64f7\u53d6\u65b9\u6cd5\uff0c\u662f\u4e00\u500b\u8457\u540d\u4e14\u6210\u6548\u975e\u5e38\u597d\u7684\u5e38\u898b\u57fa\u790e\u7cfb\u7d71\u8a2d\u7f6e\uff0c\u5728\u591a\u7a2e\u4efb\u52d9\u4e0a\u88ab\u8b49\u5be6\u80fd \u7528 0-1 \u6b0a\u91cd(\u6211\u5011\u8a8d\u70ba\u53ea\u8a2d\u5b9a\u4e00\u500b\u9580\u6abb\u503c\u5c31\u5c07\u6b0a\u91cd\u503c\u4e00\u5206\u70ba\u4e8c\u7684\u4f5c\u6cd5\u53ef\u80fd\u6703\u8f03\u7c97\u7cd9\u4e00\u4e9b)\uff1b \u6211\u5011\u5229\u7528\u5f0f(23)\u5c07E\u4e2d\u7684\u6bcf\u500b\u5143\u7d20\u4e4b\u97f3\u7d20\u932f\u8aa4\u7387\u505a\u8f49\u63db\u6b0a\u91cd\u7684\u52d5\u4f5c\uff0c\u53ef\u5c07\u6b0a\u91cd\u503c\u9650\u5236\u5728 0 \u5230 1 \u4e4b\u9593\uff1a E K=15 67.73 72.60 65.26 69.18 K=20 66.71 71.74 63.56 CMVN+HNMF 82.88 84.84 82.37 \u986f\u8457\u5730\u63d0\u5347\u8a9e\u97f3\u8fa8\u8b58\u7cfb\u7d71\u5728\u96dc\u8a0a\u74b0\u5883\u4e2d\u7684\u6548\u80fd\u3002\u7576 AFE \u8207 NSHGNMF \u505a\u7d50\u5408\u6642\u53ea\u80fd\u6709 83.56 \u4e9b\u5fae\u63d0\u5347\uff1b\u6211\u5011\u731c\u6e2c\u53ef\u80fd\u662f\u56e0\u70ba AFE \u672c\u8eab\u5df2\u5177\u5099\u6709\u5f88\u5b8c\u5584\u7684\u8a9e\u97f3\u7279\u5fb5\u6b63\u898f\u5316\u8655\u7406\u7a0b\u5e8f\uff0c 68.09 CMVN+NSGNMF 83.94 85.76 83.61 84.61 \u82e5\u518d\u52a0 NSHGNMF \u6642\uff0c\u8a9e\u97f3\u7279\u5fb5\u53ef\u80fd\u6703\u88ab\u904e\u5ea6\u5730\u6b63\u898f\u5316\u800c\u5c0e\u81f4\u8a9e\u97f3\u8fa8\u8b58\u6548\u80fd\u7121\u6cd5\u88ab\u986f\u8457 1 (23) K=30 67.43 72.66 64.05 68.84 CMVN+NSHGNMF 83.98 85.85 83.71 84.67 \u63d0\u5347\u3002 1 PER . \u7531 3.2 \u7bc0\u7684\u6558\u8ff0\u53ef\u77e5 NSNMF \u56e0\u70ba\u4e58\u6cd5\u7684\u6027\u8cea\uff0c\u82e5\u662f\u5e73\u6ed1\u7a0b\u5ea6\u9ad8\u7684\u77e9\u9663S\u8207W\u6216H\u77e9\u9663\u5176 \u4e2d\u4e00\u500b\u76f8\u4e58\uff0c\u70ba\u4e86\u8981\u88dc\u511f\u80fd\u5118\u53ef\u80fd\u7684\u8fd1\u4f3c\u91cd\u5efa\u539f\u59cb\u8cc7\u6599\uff0c\u6703\u8feb\u4f7f\u53e6\u4e00\u500b\u77e9\u9663\u9054\u5230\u7a00\u758f\u7684 \u61c9\u8a72\u662f\u5c0d\u7a31\u7684(\u4e00\u6a23\u7684)\uff1b\u56e0\u6b64\u6211\u5011\u63a1\u6298\u8877\u65b9\u5f0f\uff0c\u5c07E \u8207E \u7684\u503c\u90fd\u6539\u70ba\u5169\u8005\u7684\u76f8\u52a0\u53d6\u5e73\u5747\uff0c \u5982\u6b64\u505a\u6cd5\uff0c\u53ef\u4ee5\u5c07\u5404\u500b\u8a13\u7df4\u8a9e\u53e5\u5f7c\u6b64\u9593\u7684\u95dc\u806f\u7a0b\u5ea6\u505a\u6bd4\u8f03\u7cbe\u7d30\u7684\u63cf\u8ff0\uff0c\u800c\u4e0d\u662f\u53ea\u6709 0 \u6216 4.7 \u4e09\u7a2e\u975e\u8ca0\u77e9\u9663\u5206\u89e3\u6cd5\u6539\u9032\u65b9\u6cd5\u4e4b\u7d50\u5408 5. \u7d50\u8ad6 3.4 \u975e\u8ca0\u7de8\u78bc\u77e9\u9663\u7d71\u8a08\u5716\u7b49\u5316\u6cd5(HNMF) \u4e0d\u540c\u7684\u96dc\u8a0a\u5e72\u64fe\uff0c\u5206\u6210\u4e09\u500b\u6e2c\u8a66\u96c6\uff1aSet A\u3001Set B \u53ca Set C\u3002Set A \u7684\u8a9e\u97f3\u5206\u5225\u542b\u6709\u5730\u4e0b \u6548\u679c\u3002\u5982\u6b64\u7d93\u905e\u8ff4\u5730\u4f7f\u7528\u4e58\u6cd5\u66f4\u65b0\u898f\u5247\uff0c\u6700\u7d42\u53ef\u9054\u5230\u7a00\u758f\u5316\u77e9\u9663W\u8207H\u7684\u6548\u679c\u3002\u5be6\u9a57\u6578\u64da \u4f7f\u6b0a\u91cd\u77e9\u9663 E \u8b8a\u6210\u4e00\u500b\u5c0d\u7a31\u77e9\u9663\u3002\u518d\u8005\uff0c\u6211\u5011\u8a2d\u4e86\u4e00\u500b\u9580\u6abb\u503c(Threshold) \u03b1\uff1a 1 \u7684\u6b0a\u91cd\u503c\u3002\u4f8b\u5982\uff1a\u97f3\u7d20\u932f\u8aa4\u7387 0%\u7684\u8f49\u63db\u5f8c\u6703\u8b8a\u6210 1\uff1b\u97f3\u7d20\u932f\u8aa4\u7387 40%\u7684\u8f49\u63db\u5f8c\u6703\u8b8a\u6210 \u63a5 \u8457 \u6211 \u5011 \u7d50 \u5408 \u975e \u5e73 \u6ed1 \u975e \u8ca0 \u77e9 \u9663 \u5206 \u89e3 \u6cd5 (NSNMF) \u4ee5 \u53ca \u57fa \u65bc \u5716 \u6b63 \u5247 \u5316 \u975e \u8ca0 \u77e9 \u9663 \u5206 \u89e3 \u6cd5 Set A Set B Set C Average \u672c\u8ad6\u6587\u63a2\u8a0e\u4e86\u975e\u8ca0\u77e9\u9663\u5206\u89e3\u6cd5\u7684\u4e09\u7a2e\u6539\u9032\u65b9\u6cd5\u4e26\u5c07\u4e4b\u904b\u7528\u5728\u8a9e\u97f3\u7279\u5fb5\u7684\u8abf\u8b8a\u983b\u8b5c\u6b63\u898f\u5316 \u50b3\u7d71\u7684 NMF \u65b9\u6cd5\u5c07\u8a13\u7df4\u8cc7\u6599\u5206\u89e3\u6210\u975e\u8ca0\u57fa\u5e95\u77e9\u9663 W clean \u548c\u7de8\u78bc\u77e9\u9663 H clean \u5169\u90e8\u5206\uff0c\u5728\u6e2c \u8a66\u968e\u6bb5\u6642\u53ea\u4fdd\u7559\u57fa\u5e95\u77e9\u9663\uff0c\u800c\u4e1f\u68c4\u4e86\u7de8\u78bc\u77e9\u9663\u7684\u8cc7\u8a0a\u3002\u5728\u61c9\u7528\u4e2d\uff0c\u53d7\u566a\u97f3\u5e72\u64fe\u7684\u8a9e\u97f3\u53ef \u80fd\u6703\u5f97\u5230\u8207\u4e7e\u6de8\u8a9e\u6599\u4e0d\u76f8\u4f3c\u7684\u7de8\u78bc\u5411\u91cf\uff0c\u6b64\u6642\u6211\u5011\u4e0d\u80fd\u78ba\u5b9a\u9019\u6a23\u7684\u8cc7\u6599\u8868\u793a\u662f\u5426\u5df2\u7d93\u6392 \u9664\u5927\u90e8\u5206\u96dc\u8a0a\uff1f\u518d\u8005\uff0c\u5373\u4fbf\u662f\u4e7e\u6de8\u8a9e\u6599\u4e2d\u4e5f\u5b58\u5728\u8457\u8a31\u591a\u8b8a\u7570\u6027\u3002\u70ba\u4e86\u514b\u670d\u4e0a\u8ff0\u554f\u984c\uff0c\u672c \u8ad6\u6587\u63d0\u51fa\u5229\u7528\u7d71\u8a08\u5716\u7b49\u5316\u6cd5\u5c07\u7de8\u78bc\u77e9\u9663\u505a\u6b63\u898f\u5316\u8655\u7406\u3002\u5728\u8a13\u7df4\u968e\u6bb5\u6642\uff0c\u6211\u5011\u5229\u7528\u7d71\u8a08\u5716 \u7b49\u5316\u6cd5(HEQ)\u5c07\u4e7e\u6de8\u8a13\u7df4\u8a9e\u6599\u7684\u7de8\u78bc\u77e9\u9663 H clean \u7684\u8cc7\u8a0a\u5132\u5b58\u5efa\u8868\uff0c\u7d71\u8a08\u7de8\u78bc\u77e9\u9663 H clean \u7684 \u53c3\u8003\u5206\u5e03\uff0c\u5982\u5716 3\u3002\u800c\u5728\u6e2c\u8a66\u968e\u6bb5\u6642\u6c42\u51fa\u7de8\u78bc\u5411\u91cf h\uff0c\u518d\u5c07 h \u6bcf\u4e00\u500b\u5143\u7d20\u57f7\u884c\u7d71\u8a08\u5716\u7b49 \u5316\u6cd5\u4e4b\u67e5\u8868\u7684\u52d5\u4f5c\uff0c\u8a66\u5716\u5c07\u542b\u6709\u96dc\u8a0a\u7684 h \u9084\u539f\u56de\u5176\u5230\u5c0d\u61c9\u7684\u4e7e\u6de8\u7684\u7de8\u78bc\u5411\u91cf\u3002\u4ee5\u80fd\u5920\u53bb \u5c0d\u61c9\u7531\u4e7e\u6de8\u8a13\u7df4\u8a9e\u6599\u6240\u4f30\u6e2c\u51fa\u4f86\u7684\u53c3\u8003\u5206\u5e03\uff0c\u4f7f\u8a9e\u53e5\u7684\u7de8\u78bc\u5411\u91cf\u5728\u8a13\u7df4\u74b0\u5883\u8207\u6e2c\u8a66\u74b0\u5883 \u4e4b\u6a5f\u7387\u5206\u5e03\u4e00\u81f4\uff0c\u5982\u5716 4 \u6240\u793a\u3002\u6211\u5011\u8a8d\u70ba\u4e7e\u6de8\u7684\u57fa\u5e95\u5411\u91cf\u77e9\u9663\u4e58\u4e0a\u6b63\u898f\u5316\u5f8c\u7684\u7de8\u78bc\u77e9\u9663 \u61c9\u8f03\u80fd\u5920\u9084\u539f\u56de\u4e7e\u6de8\u7684\u8a9e\u97f3\u7279\u5fb5\u3002 \u9435(Subway)\u3001\u4eba\u8072(Babble)\u3001\u6c7d\u8eca(Car)\u548c\u5c55\u89bd\u6703\u9928(Exhibition)\u7b49\u56db\u7a2e\u52a0\u6210\u6027\u96dc\u8a0a\u8207 G.712 \u901a\u9053\u6548\u61c9\uff1bSet B \u7684\u8a9e\u97f3\u5247\u5206\u5225\u542b\u6709\u9910\u5ef3(Restaurant)\u3001\u8857\u9053(Street)\u3001\u6a5f\u5834(Airport)\u548c\u706b\u8eca \u7ad9(Train Station)\u7b49\u56db\u7a2e\u52a0\u6210\u6027\u96dc\u8a0a\u8207 G.712 \u7684\u901a\u9053\u6548\u61c9\uff1bSet C \u5206\u5225\u52a0\u5165\u4e86\u5730\u4e0b\u9435 (Subway) \u8207\u8857\u9053(Street)\u5169\u7a2e\u96dc\u8a0a\u8207 MIRS \u901a\u9053\u6548\u61c9\u3002\u800c\u5176\u4e2d\u7684\u8a0a\u566a\u6bd4(SNR)\u5247\u6709\u4e03\u7a2e\uff0c \u70ba Clean\u300120dB\u300115dB\u300110dB\u30015dB\u30010dB \u548c-5dB\uff0c\u4e26\u4e14\u63d0\u4f9b\u4e8c\u7a2e\u8a13\u7df4\u6a21\u5f0f\uff1a\u4e7e\u6de8\u60c5\u5883 \u8a13\u7df4\u6a21\u5f0f(Clean-Condition Training)\u8207\u8907\u5408\u60c5\u5883\u8a13\u7df4\u6a21\u5f0f(Multi-Condition Training)\u3002\u672c\u7814 \u7a76\u7684\u57fa\u790e\u5be6\u9a57\u7686\u4f7f\u7528\u4e7e\u6de8\u60c5\u5883\u8a13\u7df4\u6a21\u5f0f\uff0c\u6545\u5728\u8072\u5b78\u6a21\u578b\u8a13\u7df4\u6642\u4e26\u6c92\u6709\u4f7f\u7528\u5230\u4efb\u4f55\u52a0\u6210\u6027 \u96dc\u8a0a\u7684\u8cc7\u8a0a\u6216\u5167\u6db5\u3002 4.2 \u5be6\u9a57\u8a2d\u5b9a \u5728 \u672c \u8ad6 \u6587 \u4e2d \u7684 \u57fa \u790e \u5be6 \u9a57 \u662f \u63a1 \u7528 \u6885 \u723e \u5012 \u983b \u8b5c \u4fc2 \u6578 (Mel-frequency Cepstral Coefficients, MFCC)\u505a\u70ba\u8a9e\u97f3\u7279\u5fb5\u53c3\u6578\uff0c\u53d6\u6a23\u983b\u7387(Sampling Rate)\u70ba 8,000Hz\uff0c\u9810\u5f37\u8abf(Pre-emphasis) \u53c3\u6578\u8a2d\u70ba 0.97\uff1b\u4f7f\u7528\u7684\u7a97\u51fd\u6578\u70ba\u6f22\u660e\u7a97(Hamming Window)\uff0c\u97f3\u6846\u9577\u5ea6(Frame Length)\u662f 25 \u6beb\u79d2\uff0c\u97f3\u6846\u9593\u8ddd(Frame Shift)\u70ba 10 \u6beb\u79d2\u3002\u6bcf\u4e00\u500b\u97f3\u6846\u7684\u8a9e\u97f3\u7279\u5fb5\u662f\u4f7f\u7528 13 \u7dad\u6885\u723e\u5012 \u5982\u8868 1 \u6240\u793a\uff1b\u7531\u5be6\u9a57\u7d50\u679c\u53ef\u898b\uff0c\u96a8\u8457 \u7684\u589e\u52a0\u8a9e\u97f3\u8fa8\u8b58\u4e4b\u8a5e\u6b63\u78ba\u7387\u7684\u78ba\u80fd\u5920\u9010\u6f38\u5730\u88ab\u63d0 \u9ad8\u3002\u96d6\u7136\u5728 0.3\u4ee5\u524d\u8f03\u7121\u660e\u986f\u6548\u679c\uff0c\u53ef\u80fd\u56e0\u70ba\u8feb\u4f7f\u77e9\u9663\u7a00\u758f\u7684\u7a0b\u5ea6\u4e26\u4e0d\u9ad8\uff1b\u800c 1\u6642\uff0c \u8868\u793a\u8feb\u4f7f\u77e9\u9663\u7a00\u758f\u7a0b\u5ea6\u6700\u9ad8\uff0c\u800c\u6578\u64da\u986f\u793a\u7684\u78ba\u80fd\u5920\u8868\u73fe\u5f97\u6700\u597d\u3002 \u8868 1. \u975e\u5e73\u6ed1\u4e4b\u975e\u8ca0\u77e9\u9663\u5206\u89e3\u6cd5(NSNMF)\u5728\u4f7f\u7528\u4e0d\u540c\u03b8\u503c\u4e0b\u4e4b\u8a5e\u6b63\u78ba\u7387(%) Set A Set B Set C Average NMF 67.09 70.98 68.22 68.87 0.1 67.54 71.62 66.01 68.87 0.2 66.91 71.35 64.72 68.25 0.3 66.89 71.63 67.85 68.98 0.4 70.19 73.64 68.72 71.28 0.5 69.46 73.60 66.22 70.47 0.6 70.82 74.18 71.20 72.24 E \u03b1 , E 1 E , E 0 (22) \u7576E \u6216E \u5927\u65bc\u9580\u6abb\u503c\u6642\uff0c\u4ee3\u8868\u8aaa\u9019\u5169\u53e5\u8a13\u7df4\u8a9e\u53e5\u5f7c\u6b64\u9593\u7684\u97f3\u7d20\u932f\u8aa4\u7387\u8f03\u5927(\u8a13\u7df4\u8a9e\u53e5\u5dee \u7570\u5927)\uff0c\u56e0\u6b64\u95dc\u806f\u5ea6\u8a2d\u70ba 0\uff0c\u5e0c\u671b\u5169\u500b\u8a13\u7df4\u8a9e\u53e5\u5f7c\u6b64\u9593\u6c92\u6709\u95dc\u806f\uff1b\u800c\u7576E \u6216E \u5c0f\u65bc\u7b49\u65bc\u9580 \u6abb\u503c\u6642\u4ee3\u8868\u9019\u5169\u53e5\u8a13\u7df4\u8a9e\u53e5\u5f7c\u6b64\u9593\u7684\u97f3\u7d20\u932f\u8aa4\u7387\u8f03\u5c0f\uff0c\u61c9\u6709\u8f03\u5927\u7684\u95dc\u806f\u6027\uff0c\u56e0\u6b64\u8a2d\u70ba 1\u3002 0.714\uff1b\u97f3\u7d20\u932f\u8aa4\u7387 100%\u7684\u8f49\u63db\u5f8c\u6703\u8b8a\u6210 0.5\uff1b\u97f3\u7d20\u932f\u8aa4\u7387 160%\u7684\u8f49\u63db\u5f8c\u6703\u8b8a\u6210 0.385\u3002 (GNMF)\uff0c\u7a31\u70ba NSGNMF(\u4ee5\u4e0b\u6240\u7d50\u5408\u4e4b GNMF \u7686\u4f7f\u7528 GNMF-a)\uff0c\u5be6\u9a57\u6578\u64da\u5982\u8868 6 \u6240\u793a\u3002 HEQ 80.03 82.05 80.10 80.85 \u4e0a\uff1b\u5e0c\u671b\u85c9\u6b64\u80fd\u5920\u64f7\u53d6\u51fa\u66f4\u5f37\u5065\u6027\u7684\u8abf\u8b8a\u983b\u8b5c\u57fa\u5e95\u5411\u91cf\uff0c\u800c\u9054\u5230\u589e\u9032\u8a9e\u97f3\u5f37\u5065\u6027\u7684\u76ee\u7684\u3002 \u8b93\u8d8a\u4f4e\u7684\u97f3\u7d20\u932f\u8aa4\u7387\u80fd\u5920\u6709\u8d8a\u9ad8\u7684\u6b0a\u91cd\u503c\u3002\u7279\u5225\u7684\u662f\uff0c\u6b0a\u91cd\u77e9\u9663E\u4e4b\u5c0d\u6bcf\u4e00\u500b\u89d2\u7dda\u4f4d\u7f6e\u7684 \u96d6\u7136\u975e\u5e73\u6ed1\u975e\u8ca0\u77e9\u9663\u5206\u89e3\u6cd5\u539f\u672c\u5c31\u6709 78.21%\u4e4b\u4e0d\u932f\u7684\u8a5e\u6b63\u78ba\u7387\uff0c\u4e0d\u904e\u52a0\u4e0a\u57fa\u65bc\u5716\u6b63\u5247\u5316 HEQ+NSNMF 83.84 85.88 83.70 84.63 \u7b2c\u4e00\u7a2e\u662f\u975e\u5e73\u6ed1\u975e\u8ca0\u77e9\u9663\u5206\u89e3\u6cd5(NSNMF)\uff0c\u5229\u7528\u6dfb\u52a0\u4e86\u4e00\u500b\u5e73\u6ed1\u77e9\u9663 S\uff0c\u8b8a\u66f4\u50b3\u7d71\u975e\u8ca0 \u503c\uff0c\u4e5f\u5c31\u662f\u4ee3\u8868\u67d0\u4e00\u8a9e\u53e5\u672c\u8eab\u7684\u95dc\u806f\u7a0b\u5ea6\uff1b\u56e0\u5176\u97f3\u7d20\u932f\u8aa4\u7387\u70ba 0%\uff0c\u6240\u4ee5\u95dc\u806f\u7a0b\u5ea6\u6703\u70ba 1\u3002 \u76f8\u95dc\u7684\u6578\u64da\u5982\u8868 3 \u6240\u793a\uff1b\u7576\u6539\u6210\u4f7f\u7528\u5168\u57df\u90fd\u6709\u503c\u4e4b\u6b0a\u91cd\u77e9\u9663E (\u5c0d\u61c9\u65b9\u6cd5\u7c21\u7a31\u70ba GNMF-a) \u975e\u8ca0\u77e9\u9663\u5206\u89e3\u6cd5\u5229\u7528\u8a13\u7df4\u8a9e\u53e5\u9593\u7684\u76f8\u95dc\u806f\u5ea6\u4e4b\u6982\u5ff5\uff0c\u80fd\u5920\u6709 1.24%\u7684\u6b63\u78ba\u7387\u63d0\u5347\u3002\u6700\u5f8c \u6211\u5011\u518d\u5c0d\u7de8\u78bc\u77e9\u9663\u505a HEQ \u6b63\u898f\u5316\u8655\u7406\u4f86\u63d0\u5347\u8a9e\u97f3\u8fa8\u8b58\u6548\u80fd(\u8868\u793a\u6210 NSHGNMF)\uff1b\u4e0d\u904e\u518d HEQ+GNMF 83.71 84.76 82.53 \u77e9\u9663\u5206\u89e3\u6cd5\u7684\u6a21\u578b\uff1b\u5229\u7528\u6a21\u578b\u4e58\u6cd5\u7684\u6027\u8cea\uff0c\u4f7f\u4e00\u500b\u77e9\u9663\u5e73\u6ed1\uff0c\u9032\u800c\u8feb\u4f7f\u53e6\u4e00\u500b\u77e9\u9663\u9054\u5230 83.89 \u7a00\u758f\u7684\u6548\u679c\u3002\u7b2c\u4e8c\u7a2e\u662f\u57fa\u65bc\u5716\u6b63\u5247\u5316\u975e\u8ca0\u77e9\u9663\u5206\u89e3\u6cd5(GNMF)\uff0c\u5728\u6e1b\u640d\u51fd\u5f0f\u4e2d\u589e\u52a0\u4e86\u4e00\u500b \u6703\u6bd4\u4f7f\u7528\u9810\u8a2d\u7684\u4e00\u500b\u9580\u6abb\u503c\u4e4b\u6b0a\u91cd\u77e9\u9663E\u7684\u6548\u679c\u6703\u4f86\u7684\u597d\u4e00\u4e9b\uff0c\u4f7f\u8a5e\u6b63\u78ba\u7387\u63d0\u9ad8\u4e86 2.07%\uff1b \u7d50\u5408 HNMF \u4e4b\u5f8c\u6548\u679c\u4e26\u6c92\u6709\u9810\u6599\u4e2d\u986f\u8457\uff0c\u53ea\u6709\u4e9b\u8a31\u7684\u8a5e\u6b63\u78ba\u7387\u63d0\u6607\u3002\u5728\u8868 6 \u4e2d\u4e5f\u5217\u51fa\u5169 HEQ+HNMF 82.89 85.52 83.59 84.08 \u984d\u5916\u7684\u6b63\u5247\u9805\u3002\u5229\u7528\u5e7e\u4f55\u7d50\u69cb\u8207\u5c40\u90e8\u4e0d\u8b8a\u6027\u7684\u7279\u6027\uff0c\u6c42\u5f97\u8a13\u7df4\u8a9e\u53e5\u9593\u7684\u95dc\u806f\u7a0b\u5ea6\u4e26\u5275\u9020 \u540c\u6642\u4e5f\u6bd4\u50b3\u7d71\u975e\u8ca0\u77e9\u9663\u5206\u89e3\u6cd5(NMF)\u63d0\u9ad8\u4e86 3.25%\u7684\u8a5e\u6b63\u78ba\u7387\u3002 \u7a2e\u5e38\u898b\u7684\u8abf\u8b8a\u983b\u8b5c\u6b63\u898f\u5316\u6cd5(SHE \u8207 PCA)\u4f86\u4f5c\u70ba\u6bd4\u8f03\u6bd4\u8f03(Kao et al., 2014)\u3002SHE \u662f\u5229\u7528 HEQ+NSGNMF 84.02 85.89 83.79 84.72 \u4e00\u500b\u6b0a\u91cd\u77e9\u9663\u4ee5\u4f9b\u4f7f\u7528\uff0c\u4f7f\u7d93\u6b63\u898f\u5316\u7684\u8a9e\u97f3\u7279\u5fb5\u80fd\u5920\u589e\u52a0\u9451\u5225\u529b\u3002\u7b2c\u4e09\u7a2e\u662f\u975e\u8ca0\u7de8\u78bc\u77e9 \u56e0\u70ba\u8a2d\u5b9a\u4e86\u4e00\u500b\u9580\u6abb\u503c \uff0c\u82e5\u9580\u6abb\u503c \u8a2d\u5b9a\u7684\u8f03\u56b4\u683c(\u503c\u8f03\u5c0f\u6642)\uff0c\u6b0a\u91cd\u77e9\u9663 E \u5c31\u6703\u986f\u5f97\u96f6 \u503c\u8d8a\u591a\u800c\u8d8a\u7a00\u758f\u5316\u3002\u53e6\u5916\uff0c\u5728\u672c\u8ad6\u6587\u4e2d\u5c0d\u65bc\u5f0f(18)\u4e2d\u7684 \u503c\u8a2d\u5b9a\u70ba 100\u3002GNMF \u7684\u5be6\u9a57\u6578 \u503c\u5f97\u6ce8\u610f\u7684\u662f\uff0c\u5728\u4e0a\u8ff0\u5be6\u9a57\u4e2d\u4f7f\u7528\u57fa\u65bc\u97f3\u7d20\u932f\u8aa4\u7387\u6c42\u53d6\u6b0a\u91cd\u77e9\u9663E\u7684\u65b9\u5f0f\uff0c\u5728\u6b64\u65b9 \u5f0f\u4e2d\u8a9e\u97f3\u7279\u5fb5\u7684\u6240\u6709\u7dad\u5ea6\u7684\u8abf\u8b8a\u983b\u8b5c\u5f37\u5ea6\u662f\u4f7f\u7528\u76f8\u540c\u7684\u6b0a\u91cd\u77e9\u9663E\u3002\u53e6\u4e00\u65b9\u9762\uff0c\u6211\u5011\u4e5f\u5617 HEQ+NSHGNMF 84.05 85.93 83.82 \u9663\u7d71\u8a08\u5716\u7b49\u5316\u6cd5(HNMF)\uff0c\u5e0c\u671b\u80fd\u5920\u5229\u7528\u5728\u8a13\u7df4\u968e\u6bb5\u6642\u53ef\u7372\u5f97\u7684\u7de8\u78bc\u77e9\u9663\uff0c\u5229\u7528\u7d71\u8a08\u5716\u7b49 84.76 \u8868 6. NMF \u6539\u826f\u65b9\u6cd5\u7d50\u5408\u8207\u4e4b\u8a5e\u6b63\u78ba\u7387(%)\u6bd4\u8f03 \u5316\u6cd5\u5c07\u5176\u7d2f\u7a4d\u5206\u5e03\u51fd\u6578\u8cc7\u8a0a\u5efa\u8868\u5132\u5b58\uff0c\u5e0c\u671b\u5728\u6e2c\u8a66\u968e\u6bb5\u6642\u80fd\u85c9\u6b64\u5c07\u542b\u96dc\u8a0a\u8a9e\u53e5\u7684\u7de8\u78bc\u5411 \u64da\u5982\u8868 2 \u6240\u793a\u3002\u7576 \u503c\u8d8a\u9ad8\u6642\uff0c\u4ee3\u8868\u9580\u6abb\u8d8a\u5bec\u9b06\uff0c\u6b0a\u91cd\u77e9\u9663E\u4e2d\u975e\u96f6\u503c\u7684\u5143\u7d20\u4e5f\u6703\u8d8a\u591a\uff1b \u8868 2. GNMF \u4f7f\u7528\u4e0d\u540c\u9580\u6abb\u503c\u7684\u4e4b\u8a5e\u6b63\u78ba\u7387(%) Set A Set B Set C Average NMF 67.09 70.98 68.22 68.87 \u8a66\u57fa\u65bc\u8a9e\u53e5\u6bcf\u4e00\u7dad\u5ea6\u7684\u8abf\u8b8a\u983b\u8b5c\u5f37\u5ea6\uff0c\u500b\u5225\u5730\u5229\u7528\u6b50\u5f0f\u8ddd\u96e2\u4f86\u8a08\u7b97\u7684\u8a9e\u53e5\u9593\u7684\u7684\u95dc\u806f\u7a0b Set A Set B Set C \u8868 9. \u7d50\u5408 AFE \u8207 NMF \u4e4b\u8a5e\u6b63\u78ba\u7387(%) \u91cf\u9032\u4e00\u6b65\u6b63\u898f\u5316\u3002 Average \u5ea6\uff0c\u4e26\u4e14\u4e5f\u4f7f\u7528\u985e\u4f3c\u5f0f(23)\u7684\u8f49\u63db\u5f0f\uff0c\u6c42\u51fa\u4e0d\u540c\u7dad\u5ea6\u7684\u6b0a\u91cd\u77e9\u9663E\uff0c\u6b64\u65b9\u5f0f\u7c21\u7a31 GNMF-eu\u3002 NMF 67.09 70.98 68.22 68.87 Set A Set B Set C Average \u7576\u5c07\u6b64\u4e09\u7a2e\u975e\u8ca0\u77e9\u9663\u5206\u89e3\u6cd5\u4e4b\u6539\u9032\u65b9\u5f0f\u904b\u7528\u5728 Aurora-2 \u4e0a\u6642\uff0c\u7686\u80fd\u4f7f\u8a9e\u97f3\u8fa8\u8b58\u6548\u80fd \u5be6\u9a57\u7d50\u679c\u5982\u8868 4 \u6240\u793a\uff0c\u57fa\u65bc\u6b50\u5f0f\u8ddd\u96e2\u4f7f\u5f97\u4e0d\u540c\u7dad\u5ea6\u5c0d\u61c9\u8457\u4e0d\u540c\u7684\u6b0a\u91cd\u77e9\u9663\u53bb\u9032\u884c NMF \u4e2d\u77e9\u9663W \u8207 H\u6c42\u53d6\uff0c\u6700\u5f8c\u53cd\u61c9\u5728\u8a9e\u97f3\u8fa8\u8b58\u6548\u80fd\u4e0a\u4f3c\u4e4e\u6c92\u6709\u8f03\u4f7f\u7528\u97f3\u7d20\u932f\u8aa4\u7387\u7684\u65b9\u5f0f\u4f86\u7684 \u597d\u3002 GNMF 70.63 74.27 70.78 AFE 87.68 87.10 86.29 87.17 \u6709\u6240\u9032\u6b65\u3002\u6574\u9ad4\u4e0a\u4f86\u8aaa\uff0cNSNMF \u4f7f\u7528\u7684\u77e9\u9663\u7a00\u758f\u6027\u800c\u6709\u8f03\u986f\u8457\u4e14\u4e00\u81f4\u7684\u6548\u80fd\u63d0\u5347\uff1bGNMF 72.12 NSNMF 77.05 79.75 77.47 78.21 AFE+NSNMF 87.74 87.65 86.32 \u96d6\u6c92\u6709\u5e36\u4f86\u5927\u5e45\u5ea6\u7684\u6548\u80fd\u63d0\u5347\uff0c\u4f46\u662f\u5176\u6240\u5229\u7528\u8a9e\u53e5\u4e4b\u9593\u7684\u95dc\u806f\u7a0b\u5ea6\u8cc7\u8a0a\u4e5f\u80fd\u5c0d\u8a9e\u97f3\u7279\u5fb5 87.42 \u6b63\u898f\u5316\u6709\u6240\u5e6b\u52a9\uff0c\u50cf\u662f\u8207 NSNMF \u7d50\u5408\uff0c\u4e5f\u80fd\u7a0d\u5fae\u63d0\u6607\u7cbe\u78ba\u7387\uff1b\u53e6\u5916\uff0cHNMF \u5728\u5c11\u8a31\u57fa \u03b1 0.3 68.00 72.09 67.92 69.62 \u8868 4. GNMF-eu \u4e4b\u8a5e\u6b63\u78ba\u7387(%) HNMF 77.65 80.16 77.25 78.57 AFE+GNMF 87.45 87.72 86.23 87.31 \u5e95\u500b\u6578\u6642\u80fd\u63d0\u4f9b\u4e0d\u932f\u7684\u8a9e\u97f3\u8fa8\u8b58\u6548\u80fd\u63d0\u5347\u3002</td></tr><tr><td colspan=\"3\">H DH H LH \u5176\u4e2d \u662f\u7de8\u78bc\u77e9\u9663\u7684\u6b63\u5247\u9805\uff0c \u2027 \u70ba\u77e9\u9663\u7684\u8de1\u6578(Trace)\uff0cD H EH \u4f5c\u5716\u62c9\u666e\u62c9\u65af\u7b97\u5b50(Graph Laplacian)\u3002\u5728\u6b64\u5e0c\u671b\u4f7f \u6700\u5c0f\u5316\uff0c\u9054\u5230\u4fdd\u7559\u8cc7\u6599\u5c40\u90e8\u4e0d\u8b8a\u6027\u7684 \u2211 E \uff0cL D E\uff0cL\u7a31 \u5716 3. \u975e\u8ca0\u7de8\u78bc\u77e9\u9663\u7d71\u8a08\u5716\u7b49\u5316\u6cd5(HNMF)\u8a13\u7df4\u968e\u6bb5\u793a\u610f\u5716 \u983b\u8b5c\u4fc2\u6578(\u7b2c 1 \u7dad\u81f3\u7b2c 12 \u7dad\u9084\u6709\u7b2c 0 \u7dad)\uff0c\u52a0\u4e0a\u5176\u4e00\u968e\u5dee\u91cf\u548c\u4e8c\u968e\u5dee\u91cf\uff0c\u5171 39 \u7dad\u4e4b\u7279\u5fb5 0.7 72.07 75.29 69.73 72.89 \u03b1 0.5 67.86 72.22 68.02 69.64 Set A Set B Set C Average NSGNMF 78.22 80.92 78.95 AFE+HNMF 87.81 87.22 86.36 87.28 79.45 \u53c3\u6578\u3002\u672c\u8ad6\u5728\u5c0d\u8a9e\u97f3\u7279\u5fb5\u9032\u884c\u5f37\u5065\u6027(\u6b63\u898f\u5316)\u8655\u7406\u6642\uff0c\u53ea\u91dd\u5c0d 13 \u7dad\u7684\u975c\u614b\u7279\u5fb5\u53c3\u6578\u9032\u884c 0.8 72.99 76.25 72.75 74.25 \u03b1 0.7 67.07 71.18 66.77 68.65 NMF 67.09 70.98 68.22 68.87 NSGHNMF 78.28 80.96 78.98 79.49 AFE+NSGNMF 87.85 87.66 86.54 \u81f4\u8b1d 87.51 \u8655\u7406\uff0c\u5f85\u8655\u7406\u5b8c\u6210\u5f8c\u624d\u984d\u5916\u5c07\u8a9e\u97f3\u7279\u5fb5\u7684\u4e00\u968e\u5dee\u91cf\u548c\u4e8c\u968e\u5dee\u91cf\u52a0\u5165\u5f62\u6210\u6700\u5f8c\u6bcf\u4e00\u500b\u97f3\u6846 \u7684\u8a9e\u97f3\u7279\u5fb5\u3002 0.9 74.12 76.98 74.67 75.37 1 77.05 79.75 77.47 78.21 \u03b1 0.8 67.97 72.48 67.79 69.74 \u03b1 0.9 68.49 72.64 68.00 GNMF-a 70.63 74.27 70.78 72.12 SHE 74.82 77.44 76.47 AFE+NSHGNMF 87.82 87.70 86.55 87.52 \u672c\u8ad6\u6587\u4e4b\u7814\u7a76\u627f\u8499\u6559\u80b2\u90e8-\u570b\u7acb\u81fa\u7063\u5e2b\u7bc4\u5927\u5b78\u9081\u5411\u9802\u5c16\u5927\u5b78\u8a08\u756b(102J1A0800)\u8207\u884c\u653f\u9662 76.20 \u79d1\u6280\u90e8\u7814\u7a76\u8a08\u756b(MOST 104-2221-E-003-018-MY3 \u548c MOST 103-2221-E-003-016-MY2)\u4e4b 70.05 GNMF-eu 69.34 72.26 69.44 70.53 PCA 70.90 73.34 71.39 71.97 \u7d93\u8cbb\u652f\u6301\uff0c\u8b39\u6b64\u81f4\u8b1d\u3002</td></tr></table>", | |
| "num": null | |
| } | |
| } | |
| } | |
| } |