| { | |
| "paper_id": "O11-2001", | |
| "header": { | |
| "generated_with": "S2ORC 1.0.0", | |
| "date_generated": "2023-01-19T08:05:22.955131Z" | |
| }, | |
| "title": "\u6a5f\u7387\u5f0f\u8abf\u8b8a\u983b\u8b5c\u5206\u89e3\u65bc\u5f37\u5065\u6027\u8a9e\u97f3\u8fa8\u8b58 Probabilistic Modulation Spectrum Factorization for Robust Speech Recognition", | |
| "authors": [ | |
| { | |
| "first": "\u6731\u7d0b\u5100", | |
| "middle": [], | |
| "last": "\u9ad8\u4e88\u771f", | |
| "suffix": "", | |
| "affiliation": {}, | |
| "email": "" | |
| }, | |
| { | |
| "first": "\u9673\u67cf\u7433", | |
| "middle": [], | |
| "last": "\u570b\u7acb\u81fa\u7063\u5e2b\u7bc4\u5927\u5b78\u8cc7\u8a0a\u5de5\u7a0b\u5b78\u7cfb", | |
| "suffix": "", | |
| "affiliation": {}, | |
| "email": "" | |
| } | |
| ], | |
| "year": "", | |
| "venue": null, | |
| "identifiers": {}, | |
| "abstract": "", | |
| "pdf_parse": { | |
| "paper_id": "O11-2001", | |
| "_pdf_hash": "", | |
| "abstract": [], | |
| "body_text": [ | |
| { | |
| "text": "EQUATION", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [ | |
| { | |
| "start": 0, | |
| "end": 8, | |
| "text": "EQUATION", | |
| "ref_id": "EQREF", | |
| "raw_str": "(\u4e00)\u8a9e\u97f3\u7279\u5fb5\u4e4b\u8abf\u8b8a\u983b\u8b5c\u53d7\u96dc\u8a0a\u5e72\u64fe\u4e4b\u5f71\u97ff\u60c5\u5f62 \u5c0d\u65bc\u4e00\u8a9e\u97f3\u7279\u5fb5\u6642\u9593\u5e8f\u5217 ]} [ { n x \u800c\u8a00\uff0c\u5176\u8abf\u8b8a\u983b\u8b5c\u5b9a\u7fa9\u5982\u4e0b\uff1a ]) [ ( ] [ n x DFT k X \uf03d ,", | |
| "eq_num": "(1)" | |
| } | |
| ], | |
| "section": "", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "] [ k X \u8207\u539f\u59cb\u5f37\u5ea6 ] [k X \u4e4b\u95dc\u4fc2\u70ba \uf028 \uf029 ) ] [ ( ] [ 1 k X F F k X X ref \uf02d \uf03d (2) \u5176\u4e2d\uff0c ) (\uf0d7 X F \u70ba\u55ae\u4e00\u8a9e\u53e5 ]} [ { n x \u4e4b\u8abf\u8b8a\u983b\u8b5c\u5f37\u5ea6\u7684\u6a5f\u7387\u5206\u5e03\uff0c\u800c ) (\uf0d7 ref F \u5247\u70ba\u96c6\u5408\u6240\u6709\u8a13\u7df4\u8a9e \u53e5\u4e4b\u8abf\u8b8a\u983b\u8b5c\u5f37\u5ea6\u6240\u6c42\u7684\u53c3\u8003\u6a5f\u7387\u5206\u5e03\u3002 \u4e0a\u8ff0\u4e4b SHE \u6cd5\u53ef\u5c07\u7279\u5fb5\u7684\u8abf\u8b8a\u983b\u8b5c\u5f37\u5ea6\u4f5c\u975e\u7dda\u6027\u7684\u8f49\u63db\uff0c\u9032\u800c\u4f7f\u5176\u6a5f\u7387\u5206\u5e03\u6b63\u898f \u5316\uff0c\u8207\u5176\u76f8\u95dc\u9023\u7684\u65b9\u6cd5\u5305\u62ec\u4e86(\u8abf\u8b8a)\u983b\u8b5c\u5e73\u5747\u6b63\u898f\u5316\u6cd5(spectral mean normalization, SMN)[6] \u53ca \u983b \u8b5c \u5e73 \u5747 \u8207 \u8b8a \u7570 \u6578 \u6b63 \u898f \u5316 \u6cd5", | |
| "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": "PLSA \uf028 \uf029 K i T f P | \uf028 \uf029 j K s T P | \u5f37\u5ea6\u983b\u8b5c PLSA \uf028 \uf029 v | K T P \u66f4\u65b0\u4e4b\u5f37\u5ea6\u983b\u8b5c \uf028 \uf029 \uf028 \uf029 \uf0e5 \uf03d \uf0b4 K k k k i T P T f P C 1 v | | \u76f8\u4f4d\u983b\u8b5c IDFT \u66f4\u65b0\u4e4b\u7279\u5fb5\u5e8f\u5217 \u8a13\u7df4\u8207\u6e2c\u8a66\u4e4b \u7279\u5fb5\u5e8f\u5217 \u7d50\u5408\u80cc\u666f\u5f37\u5ea6\u983b\u8b5c \u7684\u5c0d\u61c9\u95dc\u4fc2\uff0c\u9996\u5148\uff0c\u6211\u5011\u5efa\u7acb\u4e00\u95dc\u4fc2\u77e9\u9663 V\uff0c\u5176\u6bcf\u4e00\u884c(column)\u662f\u500b\u5225\u8a13\u7df4\u8a9e\u53e5\u7279\u5fb5\u5e8f \u5217\u4e4b\u8abf\u8b8a\u983b\u8b5c\u5f37\u5ea6\uff0c\u9577\u5ea6\u70ba L\uff0c\u82e5\u5171\u6709 M \u53e5\u8a13\u7df4\u8a9e\u53e5\u7279\u5fb5\u5e8f\u5217\uff0c\u5247 V \u70ba M L \uf0b4 \u7684\u77e9\u9663\uff0c \u901a\u5e38 L M \uf03e\uf03e \uff0c\u63a5\u4e0b\u4f86\uff0c\u6211\u5011\u5c07\u77e9\u9663 V \u8fd1\u4f3c\u70ba\u5169\u500b\u77e9\u9663\u4e4b\u4e58\u7a4d[14]\uff1a T GH V \uf0bb (4) \u5176\u4e2d G \u8207 H T \u5206\u5225\u70ba K L \uf0b4 \u8207 M K \uf0b4 \u7684\u77e9\u9663\uff0c\u800c K \u5373\u70ba PLSA \u4e2d\u9810\u8a2d\u7684\u6f5b\u85cf\u4e3b\u984c\u500b\u6578\u3002 \u5728\u9019\u5169\u500b\u77e9\u9663\u88e1\uff1a (1) \u77e9\u9663 G \u7684\u7b2c i \u5217(row)\u7684\u5411\u91cf\uff0c\u8868\u793a\u70ba\u7b2c i \u500b\u8abf\u8b8a\u983b\u7387\u4e4b\u5f37\u5ea6\u6210\u5206(\u4ee5 i f \u8868\u793a)\u5728 \u4e0d\u540c\u6f5b\u85cf\u4e3b\u984c\u4e2d\u751f\u6210\u7684\u6a5f\u7387\u503c\uff0c (2) \u77e9\u9663 T H \u7684\u7b2c j \u884c(\u4ee5 s j \u8868\u793a)\u5247\u662f\u8868\u793a\u7b2c j \u500b\u8a9e\u53e5\u7522\u751f\u4e0d\u540c\u6f5b\u85cf\u4e3b\u984c\u7684\u4e3b\u984c\u6a5f\u7387 \u5206\u5e03\u3002\u66f4\u660e\u78ba\u5730\u8aaa\u660e\uff0c\u95dc\u4fc2\u77e9\u9663 V \u4e2d\u7684\u6bcf\u4e00\u500b\u5143\u7d20 j i a , \u88ab\u8fd1\u4f3c\u70ba\uff1a \uf028 \uf029 \uf028 \uf029 \uf028 \uf029 \uf0e5 \uf03d \uf0bb \uf03d K k j k k i j i j i s T P T f P s f P a 1 , | | | (5) \u5373\u70ba\u500b\u5225\u5e8f\u5217 j s \u900f\u904e\u6f5b\u85cf\u4e3b\u984c\u5206\u5e03\u4f30\u7b97\u7522\u751f\u8abf\u8b8a\u983b\u8b5c\u5f37\u5ea6 i f \u7684\u6a5f\u7387\u503c\u3002\u89c0\u5bdf\u4e0a\u5f0f\u53ef\u77e5\uff0c \u6a5f\u7387\u5f0f\u6f5b\u85cf\u8a9e\u610f\u5206\u6790\u6709\u5169\u5927\u985e\u7684\u6a21\u578b\u53c3\u6578\u9700\u8981\u4f30\u7b97\uff0c\u5206\u5225\u70ba\u6bcf\u4e00\u500b\u8abf\u8b8a\u983b\u8b5c\u7684\u4e3b\u984c\u6a5f\u7387 \u5206\u5e03 \uf028 \uf029 j K s T P | \u8207\u5404\u4e3b\u984c\u751f\u6210\u8abf\u8b8a\u983b\u8b5c\u5f37\u5ea6\u7684\u6a5f\u7387\u5206\u5e03 \uf028 \uf029 K i T f P | \uff1b\u800c\u9019\u4e9b\u53c3\u6578\u5247\u53ef\u7d93\u7531\u6700\u5927 \u5316\u8a13\u7df4\u8a9e\u53e5\u4e2d\u6bcf\u4e00\u500b\u8abf\u8b8a\u983b\u8b5c\u4e4b\u5c0d\u6578\u76f8\u4f3c\u5ea6(log-likelihood)\uff0c\u4e26\u4ee5\u671f\u671b\u503c\u6700\u5927\u5316\u6cd5 (expectation-maximization, EM)\u6c42\u5f97\u3002\u503c\u5f97\u4e00\u63d0\u7684\u662f\uff0c\u5df2\u6709\u7814\u7a76\u8b49\u5be6\u5728\u9069\u7576\u7684\u8a2d\u5b9a\u8207\u63a8\u5c0e \u4e0b\uff0c\u6a5f\u7387\u5f0f\u6f5b\u85cf\u8a9e\u610f\u5206\u6790\u8207\u975e\u8ca0\u77e9\u9663\u5206\u89e3(nonnegative matrix factorization, NMF)\u70ba\u7b49\u50f9 \u7684\u6982\u5ff5[15]\uff0c\u800c\u975e\u8ca0\u77e9\u9663\u5206\u89e3\u5247\u662f\u4e00\u9805\u5df2\u88ab\u5ee3\u6cdb\u904b\u7528\u65bc\u5f71\u50cf\u8655\u7406\u7684\u6f14\u7b97\u6cd5\u3002\u85c9\u7531\u6587\u737b[15] \u6240\u4ecb\u7d39\u7684\u6f14\u7b97\u6cd5\uff0c\u6211\u5011\u53ef\u4ee5\u6c42\u5f97\u4e0a\u8ff0 PLSA \u7684\u5169\u5927\u985e\u53c3\u6578 )} | ( { j K s T P \u8207 )} | ( { K i T f P \u3002 (\u4e8c)\u5229\u7528 PLSA \u751f\u6210\u6a21\u578b\u53c3\u6578\uff0c\u91cd\u5efa\u8a9e\u97f3\u7279\u5fb5\u5e8f\u5217\u4e4b\u8abf\u8b8a\u983b\u8b5c\u5f37\u5ea6 \u5728\u9019\u4e00\u6b65\u9a5f\u4e2d\uff0c\u7121\u8ad6\u662f\u6e2c\u8a66\u8a9e\u53e5\u6216\u662f\u8a13\u7df4\u8a9e\u53e5\uff0c\u5176\u539f\u59cb\u8abf\u8b8a\u983b\u8b5c\u5f37\u5ea6(\u4ee5 v \u8868\u793a)\uff0c\u7686\u7d93 \u7531\u4e0a\u4e00\u6b65\u9a5f\u6240\u5f97\u4e4b\u6a5f\u7387\u5f0f\u6f5b\u85cf\u8a9e\u610f\u5206\u6790\u7684\u5169\u500b\u6a5f\u7387\u5206\u5e03 \uf028 \uf029 K i T f P | \u8207 \uf028 \uf029 j K s T P | \uff0c\u9032\u884c\u91cd\u65b0\u4f30 \u7b97(\u65b0\u7684\u8abf\u8b8a\u983b\u8b5c\u5f37\u5ea6\uff0c\u5728\u6b64\u4ee5 \u1e7d \u8868\u793a)\u3002\u5176\u4e2d\u5fc5\u9808\u6ce8\u610f\u7684\u662f\uff0c\u4e0a\u8ff0\u4e4b \uf028 \uf029 j K s T P | \u662f\u4e7e\u6de8\u8a13\u7df4 \u7279\u5fb5\u5e8f\u5217\u5176\u8abf\u8b8a\u983b\u8b5c\u7684\u4e3b\u984c\u6a5f\u7387\u5206\u5e03\uff0c\u56e0\u6b64\u5c0d\u65bc\u6e2c\u8a66\u7279\u5fb5\u5e8f\u5217\u4e4b\u8abf\u8b8a\u983b\u8b5c v \u800c\u8a00\uff0c\u5176\u4e3b \u984c\u6a5f\u7387\u5206\u5e03\u662f\u672a\u77e5\u7684\u3002\u5728\u9019\u88e1\uff0c\u70ba\u4e86\u4f7f\u8a13\u7df4\u8a9e\u53e5\u8207\u6e2c\u8a66\u8a9e\u53e5\u7686\u901a\u904e\u76f8\u540c\u7684\u8655\u7406\u3001\u964d\u4f4e\u53ef \u80fd\u7684\u5931\u771f\uff0c\u6211\u5011\u5c0d\u65bc\u4efb\u4e00\u8abf\u8b8a\u983b\u8b5c\u5f37\u5ea6 v \uff0c\u5229\u7528\u4ee5\u4e0b\u516c\u5f0f\u9032\u884c\u4f30\u7b97\u5176\u4e3b\u984c\u6a5f\u7387\u5206\u5e03 \uf028 \uf029 v | K T P \uff1a \uf028 \uf029 \uf028 \uf029 \uf0e5 \uf0e5 \uf02b \uf03d \uf02b \uf03d \uf0d7 \uf03d 1 1 1 1 v , | v | L j j L i i k i K f f T h f T P (6) \u5176\u4e2d \uf028 \uf029 \uf0e5 \uf03d \uf03d K l l l i k k i i k T P T f P T P T f P f T h 1 ) v | ( ) | ( ) v | ( ) | ( v , | (7) \u5728\u5f97\u5230\u8abf\u8b8a\u983b\u8b5c v \u7684\u6a5f\u7387\u5206\u5e03 \uf028 \uf029 v | K T P \uff0c\u4e26\u914d\u5408\u539f\u6709\u7684 \uf028 \uf029 K i T f P | \uff0c\u6211\u5011\u5373\u53ef\u4f30\u7b97\u521d\u6b65\u66f4\u65b0 \u4e4b\u6bcf\u4e00\u7dad\u8abf\u8b8a\u983b\u8b5c i \u1e7d \uff0c\u5982\u4e0b\u5f0f\u6240\u793a\uff1a \uf028 \uf029 \uf028 \uf029 \uf0e5 \uf03d \uf0b4 \uf03d K k k k i i T P T f P C 1 v | | \u1e7d (8) \u5176\u4e2d C \u70ba\u539f\u59cb\u8abf\u8b8a\u983b\u8b5c\u5f37\u5ea6 v \u6bcf\u4e00\u7dad i v \u7684\u548c\uff0c\u5373 \uf0e5 \uf03d \uf03d L i i C 1 v \u3002 \u6b64\u5916\uff0c\u5728\u5be6\u969b\u64cd\u4f5c\u4e0a\uff0c\u7531\u65bc\u539f\u59cb\u6a5f\u7387\u5f0f\u6f5b\u85cf\u8a9e\u610f\u5206\u6790\u904b\u7528\u65bc\u8a9e\u8a00\u6a21\u578b\u6642\uff0c\u7686\u6703\u4f7f\u7528 \u6a21\u578b\u63d2\u88dc\u6cd5\uff0c\u5c07\u5176\u8207\u80cc\u666f\u6a21\u578b\u76f8\u7d50\u5408\uff1b\u5728\u6b64\u6211\u5011\u63a1\u7528\u76f8\u540c\u7684\u6982\u5ff5\uff0c\u5c07\u4e7e\u6de8\u8a9e\u97f3\u7279\u5fb5\u4e4b\u8abf \u8b8a\u983b\u8b5c\u5f37\u5ea6\u4e4b\u5e73\u5747\u4f5c\u70ba\u80cc\u666f\u8abf\u8b8a\u983b\u8b5c\u5f37\u5ea6(\u5728\u6b64\u4ee5 u \u8868\u793a)\uff0c\u518d\u5229\u7528\u63d2\u88dc\u6cd5\u5c07\u5176\u8207\u5f0f(8)\u4e4b \u1e7d \u505a\u7dda\u6027\u7d44\u5408\uff0c\u5f97\u5230\u6700\u7d42\u4e4b\u65b0\u7684\u6bcf\u4e00\u7dad\u8abf\u8b8a\u983b\u8b5c\u5f37\u5ea6 i v \uff0c\u5982\u4e0b\u5f0f\u6240\u793a\uff1a \uf028 \uf029 i i i \u1e7d 1 u v \uf061 \uf061 \uf02d \uf02b \uf03d", | |
| "eq_num": "(" | |
| } | |
| ], | |
| "section": "", | |
| "sec_num": null | |
| } | |
| ], | |
| "back_matter": [], | |
| "bib_entries": { | |
| "BIBREF0": { | |
| "ref_id": "b0", | |
| "title": "Cepstral Analysis Techniques for Automatic Speaker Verification", | |
| "authors": [ | |
| { | |
| "first": "S", | |
| "middle": [], | |
| "last": "Furui", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1981, | |
| "venue": "IEEE Trans. on Acoustic, Speech and Signal Processing", | |
| "volume": "29", | |
| "issue": "2", | |
| "pages": "254--272", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "S. Furui, \"Cepstral Analysis Techniques for Automatic Speaker Verification,\" IEEE Trans. on Acoustic, Speech and Signal Processing, Vol. 29(2): pp. 254-272, 1981", | |
| "links": null | |
| }, | |
| "BIBREF1": { | |
| "ref_id": "b1", | |
| "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": "A. Vikki, and K. Laurila, \"Segmental Feature Vector Normalization for Noise Robust Speech Recognition,\" Speech Communication, Vol. 25: pp. 133-147, 1998", | |
| "links": null | |
| }, | |
| "BIBREF2": { | |
| "ref_id": "b2", | |
| "title": "Histogram equalization of speech representation for robust speech recognition", | |
| "authors": [ | |
| { | |
| "first": "A", | |
| "middle": [ | |
| "D L" | |
| ], | |
| "last": "Torre", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "A", | |
| "middle": [ | |
| "M" | |
| ], | |
| "last": "Peinado", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "J", | |
| "middle": [ | |
| "C" | |
| ], | |
| "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 Trans. on Speech and Audio Processing", | |
| "volume": "13", | |
| "issue": "3", | |
| "pages": "355--366", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "A. D. L. Torre, A. M. Peinado, J. C. Segura, J. L. Perez-Cordoba, M. C. Benitez, and A. J. Rubio, \"Histogram equalization of speech representation for robust speech recognition,\" IEEE Trans. on Speech and Audio Processing, Vol. 13(3): pp. 355-366, 2005", | |
| "links": null | |
| }, | |
| "BIBREF3": { | |
| "ref_id": "b3", | |
| "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": "Proc. European Conf. Speech Communication and Technology", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "N. Kanedera, T. Arai, H. Hermansky, and M. Pavel, \"On the importance of various modulation frequencies for speech recognition,\" in Proc. European Conf. Speech Communication and Technology (Eurospeech), 1997", | |
| "links": null | |
| }, | |
| "BIBREF5": { | |
| "ref_id": "b5", | |
| "title": "MFCC \u7279\u5fb5\uff0c(b)\u70ba MVN \u8655\u7406\u5f8c\u4e4b\u7279\u5fb5 (c)\u70ba PLSA \u8655\u7406\u5f8c\u4e4b\u7279\u5fb5 (d)\u70ba PLSA \u7d50\u5408 MVN \u8655\u7406\u5f8c\u4e4b\u7279\u5fb5 Speech Recognition", | |
| "authors": [], | |
| "year": 2007, | |
| "venue": "Proc. IEEE Workshop on Automatic Speech Recognition and Understanding (ASRU)", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "\u5716\u4e94\u3001c1 \u7279\u5fb5\u5e8f\u5217\u65bc\u4e09\u7a2e\u8a0a\u566a\u6bd4\u60c5\u6cc1\u4e0b\u4e4b PSD \u66f2\u7dda\uff0c\u5176\u4e2d(a)\u70ba\u539f\u59cb MFCC \u7279\u5fb5\uff0c(b)\u70ba MVN \u8655\u7406\u5f8c\u4e4b\u7279\u5fb5 (c)\u70ba PLSA \u8655\u7406\u5f8c\u4e4b\u7279\u5fb5 (d)\u70ba PLSA \u7d50\u5408 MVN \u8655\u7406\u5f8c\u4e4b\u7279\u5fb5 Speech Recognition,\" in Proc. IEEE Workshop on Automatic Speech Recognition and Understanding (ASRU), 2007", | |
| "links": null | |
| }, | |
| "BIBREF6": { | |
| "ref_id": "b6", | |
| "title": "A study of sub-band modulation spectrum compensation for robust speech recognition", | |
| "authors": [ | |
| { | |
| "first": "S-Y", | |
| "middle": [], | |
| "last": "Huang", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "W", | |
| "middle": [ | |
| "H" | |
| ], | |
| "last": "Tu", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "J-W", | |
| "middle": [], | |
| "last": "Hung", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2009, | |
| "venue": "Proc. ROCLING XXI: Conf. on Computational Linguistics and Speech Processing", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "S-Y. Huang, W. H. Tu, and J-W. Hung, \"A study of sub-band modulation spectrum compensation for robust speech recognition,\" in Proc. ROCLING XXI: Conf. on Computational Linguistics and Speech Processing, 2009", | |
| "links": null | |
| }, | |
| "BIBREF7": { | |
| "ref_id": "b7", | |
| "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 Trans. on Speech and Audio Processing", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "X. Xiao, E. S. Chng, and H. Li, \"Normalization of the speech modulation spectra for robust speech recognition,\" IEEE Trans. on Speech and Audio Processing, 2008", | |
| "links": null | |
| }, | |
| "BIBREF8": { | |
| "ref_id": "b8", | |
| "title": "Constructing modulation frequency domain based features for robust speech recognition", | |
| "authors": [ | |
| { | |
| "first": "J-W", | |
| "middle": [], | |
| "last": "Hung", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "W-Y", | |
| "middle": [], | |
| "last": "Tsai", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2008, | |
| "venue": "Speech, Language Processing", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "J-W. Hung and W-Y. Tsai, \"Constructing modulation frequency domain based features for robust speech recognition,\" IEEE Trans. Acoustic, Speech, Language Processing, 2008", | |
| "links": null | |
| }, | |
| "BIBREF9": { | |
| "ref_id": "b9", | |
| "title": "RASTA processing of speech", | |
| "authors": [ | |
| { | |
| "first": "H", | |
| "middle": [], | |
| "last": "Hermansky", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "N", | |
| "middle": [], | |
| "last": "Morgan", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1994, | |
| "venue": "IEEE Trans. on Speech and Audio Processing", | |
| "volume": "2", | |
| "issue": "4", | |
| "pages": "578--589", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "H. Hermansky and N. Morgan., \"RASTA processing of speech,\" IEEE Trans. on Speech and Audio Processing, 2(4): pp. 578-589, 1994", | |
| "links": null | |
| }, | |
| "BIBREF10": { | |
| "ref_id": "b10", | |
| "title": "Integrating RASTAPLP into Speech Recognition", | |
| "authors": [ | |
| { | |
| "first": "J", | |
| "middle": [], | |
| "last": "Koehler", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1994, | |
| "venue": "Proc. IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "421--424", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "J. Koehler et al., \"Integrating RASTAPLP into Speech Recognition,\" in Proc. IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pp. 421-424, 1994", | |
| "links": null | |
| }, | |
| "BIBREF11": { | |
| "ref_id": "b11", | |
| "title": "Probabilistic latent semantic analysis", | |
| "authors": [ | |
| { | |
| "first": "T", | |
| "middle": [], | |
| "last": "Hofmann", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1999, | |
| "venue": "Proc. Uncertainty in Arterial Intelligence, UAI", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "T. Hofmann, \"Probabilistic latent semantic analysis.\" in Proc. Uncertainty in Arterial Intelligence, UAI, 1999", | |
| "links": null | |
| }, | |
| "BIBREF12": { | |
| "ref_id": "b12", | |
| "title": "Should Recognizers Have Ears?", | |
| "authors": [ | |
| { | |
| "first": "H", | |
| "middle": [], | |
| "last": "Hermansky", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1997, | |
| "venue": "Proc. ESCA-NATO Tutorial and Research Workshop on Robust speech recognition for unknown communication channels", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "1--10", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "H. Hermansky, \"Should Recognizers Have Ears?\" Invited Tutorial Paper, in Proc. ESCA-NATO Tutorial and Research Workshop on Robust speech recognition for unknown communication channels, pp. 1-10, 1997", | |
| "links": null | |
| }, | |
| "BIBREF13": { | |
| "ref_id": "b13", | |
| "title": "On the origins of speech intelligibility in the real world", | |
| "authors": [ | |
| { | |
| "first": "S", | |
| "middle": [], | |
| "last": "Greenberg", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1997, | |
| "venue": "Proc. ESCA-NATO Tutorial and Research Workshop on Robust Speech Recognition for Unknown Communication Channels", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "S. Greenberg, \"On the origins of speech intelligibility in the real world,\" in Proc. ESCA-NATO Tutorial and Research Workshop on Robust Speech Recognition for Unknown Communication Channels,1997", | |
| "links": null | |
| }, | |
| "BIBREF14": { | |
| "ref_id": "b14", | |
| "title": "Word topic models for spoken document retrieval and transcription", | |
| "authors": [ | |
| { | |
| "first": "B", | |
| "middle": [], | |
| "last": "Chen", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2009, | |
| "venue": "ACM Transaction on Asian Language Information Processing", | |
| "volume": "8", | |
| "issue": "1", | |
| "pages": "1--2", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "B. Chen, \"Word topic models for spoken document retrieval and transcription,\" ACM Transaction on Asian Language Information Processing, Vol. 8, No. 1, pp. 2:1-2:27, 2009", | |
| "links": null | |
| }, | |
| "BIBREF15": { | |
| "ref_id": "b15", | |
| "title": "Modeling vocabulary acquisition, adaptation and generalization in infants using adaptive Bayesian PLSA", | |
| "authors": [ | |
| { | |
| "first": "J", | |
| "middle": [], | |
| "last": "Driesen", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "H", | |
| "middle": [], | |
| "last": "Van Hamme", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2011, | |
| "venue": "Neurocomputing", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "J. Driesen, H. Van Hamme, \"Modeling vocabulary acquisition, adaptation and generalization in infants using adaptive Bayesian PLSA,\" Neurocomputing, 2011", | |
| "links": null | |
| }, | |
| "BIBREF16": { | |
| "ref_id": "b16", | |
| "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": "Proc. ISCA ITRW ASR", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "H. G. Hirsch and D. Pearce, \"The AURORA experimental framework for the performance evaluations of speech recognition systems under noisy conditions,\" in Proc. ISCA ITRW ASR 2000", | |
| "links": null | |
| }, | |
| "BIBREF17": { | |
| "ref_id": "b17", | |
| "title": "MVA processing of speech features", | |
| "authors": [ | |
| { | |
| "first": "C-P", | |
| "middle": [], | |
| "last": "Chen", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "J", | |
| "middle": [], | |
| "last": "Bilmes", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2007, | |
| "venue": "IEEE Trans. on Audio, Speech and Language Processing", | |
| "volume": "15", | |
| "issue": "1", | |
| "pages": "257--269", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "C-P. Chen and J. Bilmes, \"MVA processing of speech features,\" IEEE Trans. on Audio, Speech and Language Processing, Vol. 15, No. 1, pp. 257-269, 2007", | |
| "links": null | |
| }, | |
| "BIBREF18": { | |
| "ref_id": "b18", | |
| "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": "Proc. APSIPA Annual Summit and Conference (APSIPA ASC)", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "W-Y. Chu, J-W. Hung and B. Chen, \"Modulation spectrum factorization for robust speech recognition,\" in Proc. APSIPA Annual Summit and Conference (APSIPA ASC), 2011", | |
| "links": null | |
| }, | |
| "BIBREF19": { | |
| "ref_id": "b19", | |
| "title": "Robust speech recognition using spatial-temporal feature distribution characteristics", | |
| "authors": [ | |
| { | |
| "first": "B", | |
| "middle": [], | |
| "last": "Chen", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "W-H", | |
| "middle": [], | |
| "last": "Chen", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "S-H", | |
| "middle": [], | |
| "last": "Lin", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "W-Y.", | |
| "middle": [], | |
| "last": "Chu", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2011, | |
| "venue": "Pattern Recognition Letters", | |
| "volume": "32", | |
| "issue": "7", | |
| "pages": "919--926", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "B. Chen, W-H. Chen, S-H. Lin, and W-Y. Chu, \"Robust speech recognition using spatial-temporal feature distribution characteristics,\" Pattern Recognition Letters, Vol. 32, No. 7, pp. 919-926, 2011", | |
| "links": null | |
| } | |
| }, | |
| "ref_entries": { | |
| "TABREF2": { | |
| "text": "(spectral mean and variance normalization, SMVN)[6]\u7b49\u3002\u9019\u5169\u7a2e\u65b9\u6cd5\u5229\u7528\u4e86\u7dda\u6027\u8f49\u63db(linear transform)\uff0c\u5206\u5225\u5c0d\u8abf\u8b8a\u983b\u8b5c\u5f37\u5ea6\u4e4b\u5e73 \u5747\u503c\u3001\u6216\u5e73\u5747\u503c\u8207\u8b8a\u7570\u6578\u52a0\u4ee5\u6b63\u898f\u5316\uff0c\u985e\u4f3c SHE \u7684\u89c0\u5ff5\uff0cSMN \u8207 SMVN \u540c\u6a23\u53ef\u5c07\u4e7e \u6de8\u8a9e\u97f3\u7279\u5fb5\u8207\u96dc\u8a0a\u8a9e\u97f3\u7279\u5fb5\u4e4b\u8abf\u8b8a\u983b\u8b5c\u5f37\u5ea6\u4e4b\u9593\u7684\u4e0d\u5339\u914d\u964d\u4f4e\uff0c\u9032\u800c\u63d0\u5347\u7279\u5fb5\u7684\u96dc\u8a0a\u5f37 \u5065\u6027\u3002 2\u3001\u5206\u983b\u6bb5\u8abf\u8b8a\u983b\u8b5c\u7d71\u8a08\u6b63\u898f\u5316\u6cd5 \u524d\u9762\u6240\u63d0\u5230\u7684 SHE, SMN \u8207 SMVN \u4e09\u7a2e\u65b9\u6cd5\uff0c\u662f\u5c07\u5168\u90e8\u8abf\u8b8a\u983b\u5e36\u4e4b\u983b\u8b5c\u5f37\u5ea6\u503c\u8996\u70ba\u540c \u4e00\u96a8\u6a5f\u8b8a\u6578(random variable)\u7684\u6a23\u672c(samples)\uff0c\u9032\u800c\u4e00\u9f4a\u4f5c\u6b63\u898f\u5316\u3002\u7136\u800c\uff0c\u5982\u524d\u6240\u8ff0\uff0c\u4e0d", | |
| "content": "<table><tr><td/><td/><td colspan=\"5\">\u8a13\u7df4\u4e4b\u7279\u5fb5\u5e8f\u5217</td><td/></tr><tr><td/><td/><td/><td/><td/><td>DFT</td><td>DFT</td><td/></tr><tr><td colspan=\"5\">\u76f8\u4f4d\u983b\u8b5c</td><td/><td/><td/></tr><tr><td/><td/><td colspan=\"5\">\u8a13\u7df4\u96c6\u4e4b\u5f37\u5ea6\u983b\u8b5c</td><td/></tr><tr><td colspan=\"7\">\u5716\u4e8c\u3001\u4ee5\u6a5f\u7387\u5f0f\u6f5b\u85cf\u8a9e\u610f\u5206\u6790\u70ba\u57fa\u790e\u4e4b\u8abf\u8b8a\u983b\u8b5c\u6b63\u898f\u5316\u6cd5\u4e4b\u7a0b\u5e8f</td><td/></tr><tr><td>H</td><td>[</td><td>k</td><td>]</td><td>\uf03d</td><td>XX P P SS</td><td>] ] k k [ [</td><td>(3)</td></tr><tr><td colspan=\"8\">STEP 3:\u5c07\u5f0f(3)\u505a\u53cd\u96e2\u6563\u5085\u7acb\u8449\u8f49\u63db(IDFT)\uff0c\u6240\u5f97\u7684\u5e8f\u5217\u5148\u5f8c\u7d93\u904e\u7a97\u5316(windowing)\u8207\u76f4</td></tr><tr><td colspan=\"8\">\u6d41\u589e\u76ca(DC gain)\u6b63\u898f\u5316\u5f8c\uff0c\u6700\u5f8c\u6240\u5f97\u7684\u5e8f\u5217\u5373\u70baTSN\u6240\u7528\u7684\u6ffe\u6ce2\u5668\u4e4b\u8108\u885d\u97ff\u61c9(impulse \u540c\u8abf\u8b8a\u983b\u7387\u7684\u6210\u5206\u5728\u8a9e\u97f3\u8fa8\u8b58\u4e2d\u5b58\u5728\u4e0d\u7b49\u50f9\u7684\u91cd\u8981\u6027\uff0c\u4f4e\u983b\u6210\u5206\u6bd4\u9ad8\u983b\u6210\u5206\u76f8\u5c0d\u91cd\u8981\u3002 response)\uff0c\u4ee5 ]} [ { n h \u8868\u793a\u3002 \u56e0\u6b64\uff0c\u6587\u737b[6]\u63d0\u51fa\u5c07\u8abf\u8b8a\u983b\u5e36\u5207\u5272\u6210\u591a\u6bb5\u7684\u5b50\u983b\u6bb5\uff0c\u518d\u5206\u5225\u5c0d\u6bcf\u4e00\u500b\u5b50\u983b\u6bb5\u7684\u983b\u8b5c\u5f37 \u503c\u5f97\u6ce8\u610f\u7684\u662f\uff0c\u4e0a\u8ff0\u7684\u6ffe\u6ce2\u5668\u983b\u7387\u97ff\u61c9 ]} [ { n h \u662f\u96a8\u4e0d\u540c\u7279\u5fb5\u5e8f\u5217\u800c\u6539\u8b8a(\u56e0\u70ba\u5f0f(3)\u88e1\u7684 \u5ea6\u4f5c\u7d71\u8a08\u503c(\u5982\u5148\u524d\u6240\u63d0\u7684\u5e73\u5747\u503c\u3001\u8b8a\u7570\u6578\u6216\u7d71\u8a08\u5716)\u6b63\u898f\u5316\u8655\u7406\uff0c\u800c\u70ba\u4e86\u5f37\u8abf\u8f03\u4f4e\u8abf\u8b8a ]} [ { k P XX \u662f\u500b\u5225\u7279\u5fb5\u5e8f\u5217\u7684PSD)\uff0c\u500b\u5225\u7279\u5fb5\u5e8f\u5217\u901a\u904e\u5176\u5c0d\u61c9\u7684TSN\u6ffe\u6ce2\u5668\u5f8c\uff0c\u65b0\u7279\u5fb5\u5e8f \u983b\u7387\u7684\u91cd\u8981\u6027\uff0c\u5728\u4f4e\u983b\u90e8\u5206\uff0c\u5b50\u983b\u6bb5\u7684\u983b\u5bec\u8f03\u7d30\u3001\u5b50\u983b\u6bb5\u7684\u6578\u76ee\u8f03\u591a\uff0c\u9ad8\u983b\u90e8\u5206\u5247\u662f\u76f8 \u5217\u7684PSD\u6703\u903c\u8fd1\u65bc\u53c3\u8003PSD\uff0c\u7531\u65bcPSD\u53ef\u8996\u70ba\u5e73\u7de9\u5316\u5f8c(smoothed)\u7684\u8abf\u8b8a\u983b\u8b5c\u5f37\u5ea6\u5e73\u65b9\uff0c \u53cd\u3002\u6839\u64da\u6587\u737b[6]\u7684\u5be6\u9a57\u6578\u64da\u986f\u793a\uff0c\u5206\u983b\u6bb5\u6b63\u898f\u5316\u76f8\u5c0d\u65bc\u5168\u983b\u5e36\u6b63\u898f\u5316\u800c\u8a00\uff0c\u53ef\u4ee5\u5f97\u5230 \u66f4\u4f73\u7684\u8fa8\u8b58\u7387\uff0c\u7136\u800c\uff0c\u5176\u8a08\u7b97\u8907\u96dc\u5ea6\u8207\u6240\u9700\u8a18\u61b6\u9ad4\u7a7a\u9593\u4e5f\u8f03\u5927\u3002 3\u3001\u6642\u9593\u5e8f\u5217\u7d50\u69cb\u6b63\u898f\u5316\u6cd5(TSN) \u6642\u9593\u5e8f\uf99c\u7d50\u69cb\u6b63\u898f\u5316\u6cd5(temporal structure normalization, TSN)[7] \u662f\u5c6c\u65bc\u4e00\u7a2e\u6642\u9593\u5e8f\uf99c \uf984\u6ce2\u5668(temporal filter)\u8a2d\u8a08\u4e4b\u6280\u8853\uff0c\u5176\u85c9\u7531\u8a9e\u97f3\u7279\u5fb5\u5e8f\uf99c\u901a\u904e\u4e00\u4e8b\u5148\u8a2d\u8a08\u4e4b\uf984\u6ce2\u5668\uff0c\u4ee5 \u9054\u5230\u6b63\u898f\u5316\u8abf\u8b8a\u983b\u8b5c\u4e4b\u76ee\u7684\u3002\u8332\u5c07TSN\u6cd5\u6240\u4f7f\u7528\u7684\u6ffe\u6ce2\u5668\u8a2d\u8a08\u6b65\u9a5f\u7c21\u8ff0\u5982\u4e0b\uff1a STEP 1: \u5c07\u8a13\u7df4\u8a9e\u6599\u5eab\u4e2d\uff0c\u6240\u6709\u4e7e\u6de8\u8a9e\u97f3\u7279\u5fb5\u5e8f\u5217(\u5c0d\u55ae\u4e00\u7a2e\u985e\u4e4b\u7279\u5fb5\u800c\u8a00)\u5c0d\u61c9\u4e4b\u529f\u7387 \u983b\u8b5c\u5bc6\u5ea6(power spectral density, PSD)\u4f5c\u5e73\u5747\uff0c\u6b64\u5e73\u5747\u8996\u70ba\u53c3\u8003\u529f\u7387\u983b\u8b5c\u5bc6\u5ea6\uff0c\u4ee5 ]} [ { k \u6545TSN\u7684\u76ee\u6a19\u76f8\u7576\u65bc\u5c07\u8a9e\u97f3\u7279\u5fb5\u5e8f\u5217\u7684\u8abf\u8b8a\u983b\u8b5c\u5f37\u5ea6\u4e00\u81f4\u5316\uff0c\u85c9\u4ee5\u964d\u4f4e\u56e0\u96dc\u8a0a\u5e72\u64fe\u5728\u8abf \u8b8a\u983b\u8b5c\u5f37\u5ea6\u9020\u6210\u7684\u8b8a\u7570\u3002 P SS \u8868\u793a\uff0c\u5176\u4e2dk\u70ba\u983b\u7387\u7d22\u5f15\u3002 STEP 2: \u5c0d\u8a13\u7df4\u8207\u6e2c\u8a66\u8a9e\u6599\u5eab\u4e2d\uff0c\u6c42\u53d6\u500b\u5225\u8a9e\u97f3\u7279\u5fb5\u5e8f\u5217\u4e4b\u529f\u7387\u983b\u8b5c\u5bc6\u5ea6\uff0c\u4ee5 ]} [ { k \u4e09\u3001\u4ee5\u6a5f\u7387\u5f0f\u6f5b\u85cf\u8a9e\u610f\u5206\u6790\u70ba\u57fa\u790e\u4e4b\u8abf\u8b8a\u983b\u8b5c\u6b63\u898f\u5316\u6cd5 \u672c\u8ad6\u6587\u5617\u8a66\u6a5f\u7387\u5f0f\u6f5b\u85cf\u8a9e\u610f\u5206\u6790(probabilistic latent semantic analysis, PLSA)[11]\u61c9\u7528\u65bc \u8abf\u8b8a\u983b\u8b5c\u8655\u7406\uff0c\u5176\u662f\u4e00\u7a2e\u4f7f\u7528\u6a5f\u7387\u6a21\u578b\u7684\u65b9\u5f0f\uff0c\u627e\u51fa\u8abf\u8b8a\u983b\u8b5c\u5f37\u5ea6\u8207\u4e0d\u540c\u8a9e\u97f3\u7279\u5fb5\u5e8f\u5217 \u4e4b\u9593\u7684\u4e3b\u984c\u8cc7\u8a0a\u3002PLSA \u53ef\u88ab\u8996\u70ba\u662f\u4e00\u7a2e\u89c0\u9ede\u6a21\u578b(aspect model)\u7684\u5206\u6790\uff0c\u5176\u900f\u904e\u4e00\u7d44\u96b1 \u85cf\u8b8a\u6578\u7684\u6a5f\u7387\u5206\u5e03\uff0c\u4ee5\u5171\u540c\u9810\u6e2c\u4e00\u4e8b\u4ef6\u767c\u751f\u7684\u53ef\u80fd\u6027\uff0c\u800c\u6b64\u7d44\u96b1\u85cf\u8b8a\u6578\uff0c\u5373\u53ef\u88ab\u55bb\u70ba\u4e00 \u7d44\u6f5b\u85cf\u4e3b\u984c\u3002\u7576\u6211\u5011\u4f7f\u7528 PLSA \u4f86\u66f4\u65b0\u8a9e\u97f3\u7279\u5fb5\u6642\u9593\u5e8f\u5217\u7684\u8abf\u8b8a\u983b\u8b5c\u5f37\u5ea6\u6642\uff0c\u5176\u6d41\u7a0b\u5716 P XX \u8868 \u5982\u5716\u4e8c\u6240\u793a\uff0c\u800c\u8a73\u7d30\u6b65\u9a5f\u9673\u8ff0\u5982\u4e0b\uff1a</td></tr><tr><td colspan=\"7\">\u793a\uff0c\u5247\u6ffe\u6ce2\u5668\u7684\u983b\u7387\u97ff\u61c9(frequency response)\u5b9a\u70ba\uff1a (\u4e00)\u85c9\u7531\u4e7e\u6de8\u8a9e\u97f3\u7279\u5fb5\u5e8f\u5217\u4e4b\u8abf\u8b8a\u983b\u8b5c\u5f37\u5ea6\uff0c\u6c42\u53d6\u5176\u5c0d\u61c9\u7684 PLSA \u751f\u6210\u6a21\u578b</td><td/></tr><tr><td colspan=\"8\">\u6211\u5011\u4f7f\u7528 PLSA \u7684\u89c0\u5ff5\uff0c\u70ba\u6bcf\u4e00\u53e5\u4e7e\u6de8\u8a13\u7df4\u8a9e\u53e5\u4e4b\u7279\u5fb5\u5e8f\u5217\u7684\u8abf\u8b8a\u983b\u8b5c\u5f37\u5ea6\u5efa\u7acb\u751f\u6210\u6a21</td></tr><tr><td colspan=\"8\">\u578b\uff0c\u5176\u900f\u904e\u4e00\u7d44\u5171\u4eab\u7684\u6f5b\u85cf\u4e3b\u984c\u6a5f\u7387\u5206\u5e03\uff0c\u4ee5\u63cf\u7e6a\u6bcf\u4e00\u8a9e\u97f3\u7279\u5fb5\u5e8f\u5217\u8207\u5176\u8abf\u8b8a\u983b\u8b5c\u5f37\u5ea6</td></tr></table>", | |
| "type_str": "table", | |
| "num": null, | |
| "html": null | |
| }, | |
| "TABREF3": { | |
| "text": "\u4e2d\u77e9\u9663 G \u7684\u884c\u5411\u91cf)\uff0c\u5716\u4e09(a)\u662f\u5c0d\u61c9\u539f\u59cb MFCC \u4e4b c1 \u7279\u5fb5\uff0c\u5716\u4e09 (b)\u5247\u662f\u5c0d\u61c9\u7d93 MVN \u8655\u7406\u5f8c MFCC \u4e4b c1 \u7279\u5fb5\u3002\u9019\u5169\u5716\u90fd\u986f\u793a\u4e86\uff0cPLSA \u6240\u5f97\u4e4b\u6f5b\u85cf\u4e3b \u5716\u4e09\u3001PLSA \u5c0d\u65bc(a)\u539f\u59cb MFCC \u4e4b c1 (b)MVN \u8655\u7406\u5f8c MFCC \u4e4b c1 \u6240\u6c42\u5f97\u7684\u4e94\u500b\u4e3b\u984c \u6a5f\u7387\u5206\u5e03\u983b\u8b5c\u5f37\u5ea6 \u984c\u8abf\u8b8a\u983b\u8b5c\u5f37\u5ea6\u90fd\u96c6\u4e2d\u5728\u4f4e\u983b\u6210\u5206(\u5927\u7d04 10 Hz \u4ee5\u4e0b)\uff0c\u5982\u524d\u6240\u8ff0\uff0c\u9019\u5340\u57df\u6b63\u662f\u91cd\u8981\u8a9e\u97f3 \u8cc7\u8a0a\u532f\u96c6\u4e4b\u8655\uff0c\u986f\u793a\u4e86 \u76f8\u5c0d\u65bc\u4f7f\u7528\u539f\u59cb MFCC \u7279\u5fb5\u4e4b\u57fa\u790e\u5be6\u9a57\u800c\u8a00\uff0cCMS \u8207 MVN \u7686\u80fd\u6539\u5584\u8fa8\u8b58\u7cbe\u78ba\u7387\uff0c \u5176\u4e2d\u53c8\u4ee5 MVN \u7684\u6539\u9032\u6548\u679c\u8f03\u597d\uff0c\u53ef\u63d0\u4f9b\u9ad8\u9054 17%\u5de6\u53f3\u7684\u7cbe\u78ba\u7387\u63d0\u5347\u3002\u800c\u6211\u5011\u6240\u63d0\u7684 PLSA \u6cd5\uff0c\u5728\u56db\u7a2e\u6f5b\u5728\u4e3b\u984c\u6578\u7684\u9078\u64c7\u4e0b\uff0c\u7686\u512a\u65bc MVN \u6cd5\u3002 (2) \u7576 PLSA \u6cd5\u8207 CMS \u7d50\u5408\u6642\uff0c\u76f8\u8f03\u65bc\u55ae\u4e00 PLSA \u6cd5\u6216\u55ae\u4e00 CMS \u6cd5\u800c\u8a00\uff0c\u90fd\u80fd\u4f7f\u8fa8\u8b58 \u7cbe\u78ba\u7387\u66f4\u6709\u6548\u7684\u63d0\u5347\uff0c\u6b64\u9032\u6b65\u7684\u73fe\u8c61\u4e5f\u540c\u6a23\u767c\u751f\u65bc PLSA \u6cd5\u8207 MVN \u7684\u7d50\u5408\u4e0a\u3002\u6574\u9ad4\u5e73 \u5747\u8fa8\u8b58\u7387\u90fd\u53ef\u8d85\u904e 90%\uff0c\u53e6\u5916\uff0c\u5728\u7d50\u5408 PLSA \u6cd5\u7684\u524d\u63d0\u4e0b\uff0cCMS \u8207 MVN \u7684\u8868\u73fe\u5dee\u8ddd \u5f88\u5c0f(\u8fa8\u8b58\u7cbe\u78ba\u7387\u7684\u5dee\u8ddd\u50c5\u6709 0.5%\u5de6\u53f3)\uff0c \u9019\u4ee3\u8868\u4e86\u5728 PLSA \u6cd5\u524d\u7f6e\u8655\u7406\u65b9\u6cd5\u7684\u9078\u64c7\u4e0a\uff0c \u6211\u5011\u53ef\u4f7f\u7528\u7c21\u6613\u7684 CMS \u6cd5\uff0c\u5373\u53ef\u8da8\u65bc\u8f03\u8907\u96dc\u7684 MVN \u9054\u5230\u7684\u6548\u80fd\u3002 (3) \u8ddf\u8868\u4e00\u5448\u73fe\u7684\u6578\u64da\u985e\u4f3c\uff0c\u5728\u7d50\u5408 CMS \u6216 MVN \u5f8c\u7684 PLSA\uff0c\u5176\u6f5b\u5728\u4e3b\u984c\u6578\u76ee\u7684\u591a\u5be1 \u8207\u8fa8\u8b58\u7cbe\u78ba\u7387\u4e26\u7121\u986f\u8457\u7684\u95dc\u4fc2\u3002\u800c\u6539\u8b8a\u6f5b\u5728\u4e3b\u984c\u6578\u76ee\u6240\u9020\u6210\u7684\u5e73\u5747\u7cbe\u78ba\u7387\u8b8a\u5316\u7686\u5728 0.3%\u4ee5\u4e0b\uff0c\u9019\u4e5f\u986f\u793a\u4e86\u6211\u5011\u53ef\u4ee5\u4f7f\u7528\u5f88\u5c11\u7684\u6f5b\u5728\u4e3b\u984c(\u5982 K=5)\uff0c\u4e5f\u5c31\u662f\u8aaa\u5728\u7c21\u5316\u904b\u7b97\u8907 \u96dc\u5ea6\u7684\u524d\u63d0\u4e0b\u4ea6\u4e0d\u5f71\u97ff PLSA \u6cd5\u7684\u512a\u7570\u6027\u3002 \u8868\u4e8c\u3001 PLSA \u70ba\u57fa\u790e\u4e4b\u65b9\u6cd5\u4f5c\u7528\u65bc\u7d93 CMS \u8655\u7406\u4e4b MFCC \u7279\u5fb5\u7684\u8fa8\u8b58\u7d50\u679c\uff0c\u5176\u4e2d RR 1 (%)\u8207 RR 2 (%)\u5206\u5225\u70ba\u5c0d\u6bd4\u65bc\u57fa\u790e\u5be6\u9a57\u8207 CMS \u6cd5\u4e4b\u8fa8\u8b58\u7387\u4e4b\u76f8\u5c0d\u932f\u8aa4\u964d\u4f4e\u7387\u3002 \u5e73\u5747\u8a5e\u7cbe\u78ba\u7387 SHE \u7b49\uff0c\u90fd\u662f\u76f4\u63a5\u6216\u9593\u63a5\u5730\u66f4\u65b0\u7279\u5fb5\u4e4b\u8abf\u8b8a\u983b\u8b5c\uff0c\u9032\u800c\u5f37\u5316\u96dc\u8a0a\u5f37\u5065\u6027\u3002\u7531\u65bc\u5728\u6587\u737b\u4e2d\u63d0\u53ca \u9019\u4e9b\u6280\u8853\u6642\uff0c\u90fd\u662f\u76f4\u63a5\u5c55\u793a\u5176\u4f5c\u7528\u65bc MVN \u524d\u7f6e\u8655\u7406\u5f8c\u4e4b\u7279\u5fb5\u7684\u8fa8\u8b58\u6548\u679c\uff0c\u5728\u9019\u88e1\uff0c\u6211 \u5011\u540c\u6a23\u5148\u5c07\u539f\u59cb\u8a9e\u97f3\u7279\u5fb5 MFCC \u5148\u7d93 MVN \u8655\u7406\u5f8c\uff0c\u518d\u5206\u5225\u904b\u4f5c\u9019\u4e9b\u6280\u8853\u3002 \u5716\u56db\u4e2d\u5c55\u793a\u4e86\u4e0a\u8ff0\u9019\u4e9b\u6280\u8853\u6240\u5f97\u4e4b\u5e73\u5747\u8fa8\u8b58\u7cbe\u78ba\u7387\uff0c\u6211\u5011\u63d0\u51fa\u7684 PLSA \u6cd5\u6240\u63a1\u7528\u7684 \u96b1\u85cf\u4e3b\u984c\u6578\u70ba 20\u3002\u5f9e\u6b64\u5716\u4e2d\uff0c\u6211\u5011\u770b\u5230\u9019\u88e1\u6240\u7528\u7684\u6240\u6709\u65b9\u6cd5\u7686\u80fd\u63d0\u5347 MVN \u7279\u5fb5\u7684\u8fa8\u8b58 \u7cbe\u78ba\u7387\uff0c\u6240\u5f15\u7528\u7684\u4e09\u7a2e\u65b9\u6cd5\u4e2d\uff0c\u53c8\u4ee5 TSN \u7684\u6548\u80fd\u6700\u597d\uff0c\u80fd\u9054\u5230 91.02%\u7684\u7e3d\u5e73\u5747\u8fa8\u8b58\u7387\uff0c \u96d6\u7136\u6211\u5011\u63d0\u51fa\u7684 PLSA \u6cd5\uff0c\u8fa8\u8b58\u6548\u80fd\u7565\u4f4e\u65bc TSN\uff0c\u4f46\u4e5f\u53ef\u4f7f\u7e3d\u5e73\u5747\u8fa8\u8b58\u7387\u63d0\u5347\u81f3 90.57%\uff0c\u6b64\u521d\u6b65\u986f\u793a\u4e86 PLSA \u6cd5\u8db3\u4ee5\u8207\u73fe\u4eca\u6709\u540d\u7684\u8abf\u8b8a\u983b\u8b5c\u66f4\u65b0\u6280\u8853\u5728\u6548\u80fd\u4e0a\u4e26\u99d5\u9f4a \u9a45\u3002 4\u3001PLSA \u964d\u4f4e\u8abf\u8b8a\u983b\u8b5c\u5f37\u5ea6\u5931\u771f\u7684\u6548\u80fd \u6700\u5f8c\uff0c\u9664\u4e86\u8fa8\u8b58\u7cbe\u78ba\u7387\u7684\u9a57\u8b49\uff0c\u6211\u5011\u5617\u8a66\u9032\u4e00\u6b65\u85c9\u7531 PLSA \u6cd5\u65bc\u4e0d\u540c\u8a0a\u566a\u6bd4(SNR)\u4e0b\u6240 \u5f97\u4e4b\u7279\u5fb5\u5e8f\u5217\u8abf\u8b8a\u983b\u8b5c\u5f37\u5ea6\uff0c\u6aa2\u8996\u5176\u964d\u4f4e\u96dc\u8a0a\u6240\u7522\u751f\u4e4b\u5931\u771f\u7684\u80fd\u529b\u3002\u5716\u4e94(a)-(d)\u70ba\u55ae\u4e00 \u8a9e\u53e5\u5176\u539f\u59cb\u8207\u7d93\u904e\u5404\u7a2e\u8655\u7406\u65b9\u6cd5\u5f8c\u4e4b\u7b2c\u4e00\u7dad\u6885\u723e\u5012\u983b\u8b5c\u7279\u5fb5\u53c3\u6578 c1 \u65bc\u4e09\u7a2e\u4e0d\u540c\u8a0a\u566a\u6bd4 (clean\u300110dB \u8207 0dB)\u4e4b\u529f\u7387\u983b\u8b5c\u5bc6\u5ea6(power spectral density, PSD)\u3002\u9996\u5148\uff0c\u89c0\u5bdf\u5716\u4e94(a) \u53ef\u767c\u73fe\uff0c\u96dc\u8a0a\u5e72\u64fe\u6240\u5f15\u767c\u7684\u4e0d\u5339\u914d\u6548\u61c9\uff0c\u660e\u986f\u666e\u904d\u5b58\u5728\u6574\u500b\u8abf\u8b8a\u983b\u7387\u7bc4\u570d[0, 50Hz]\u5167\u3002 \u5716\u4e94(b)\u5247\u986f\u793a\uff0cMVN \u53ef\u6709\u6548\u964d\u4f4e\u4f4e\u8abf\u8b8a\u983b\u7387\u4e4b PSD \u5931\u771f\uff0c\u4f46\u662f\u4e0d\u5339\u914d\u60c5\u5f62\u4ecd\u820a\u5b58\u5728\u65bc \u4e2d\u9ad8\u8abf\u8b8a\u983b\u7387\u6210\u5206\u3002\u5716\u4e94(c)\u8207\u5716\u4e94(d)\u5247\u662f\u524d\u8ff0\u5716\u4e94(a)\u8207(b)\u5169\u985e\u7279\u5fb5\u53c3\u6578\u7d93\u904e PLSA \u8655 \u7406\u904e\u5f8c\u4e4b PSD \u66f2\u7dda\uff1b\u5f9e\u9019\u5169\u5716\u7686\u53ef\u767c\u73fe\uff0c\u85c9\u7531 PLSA \u6cd5\uff0c\u53ef\u5927\u5e45\u964d\u4f4e\u6574\u500b\u983b\u7387\u7bc4\u570d\u7684 PSD \u5931\u771f\u3002\u800c\u76f8\u8f03\u65bc\u5716\u4e94(c) \uff0c\u5716\u4e94(d)\u4e2d\u65bc\u4e0d\u540c\u8a0a\u566a\u6bd4\u4e4b PSD \u66f2\u7dda\u5247\u66f4\u70ba\u4e00\u81f4\uff0c\u986f\u793a PLSA \u8207 MVN \u7d50\u5408\u5f8c\uff0c\u80fd\u66f4\u6709\u6548\u964d\u4f4e\u96dc\u8a0a\u7522\u751f\u7684\u5931\u771f\u3002", | |
| "content": "<table><tr><td>\u8868\u4e00\u3001 PLSA \u6cd5\u4f5c\u7528\u65bc\u539f\u59cb MFCC \u7279\u5fb5\u7684\u8fa8\u8b58\u7d50\u679c\uff0c\u5176\u4e2d Avg(%)\u8207 RR(%)\u5206\u5225\u70ba\u7e3d</td></tr><tr><td>9) \u6700\u5f8c\uff0c\u6211\u5011\u5c07\u66f4\u65b0\u5f8c\u4e4b\u8abf\u8b8a\u983b\u8b5c\u5f37\u5ea6\u8207\u539f\u59cb\u8abf\u8b8a\u983b\u8b5c\u76f8\u4f4d\u505a\u7d44\u5408\uff0c\u4e26\u7d93\u7531\u53cd\u5085\u7acb\u8449 \u5176\u4e2d\uf061 \u70ba\u52a0\u6b0a\u503c\u3002 \u8f49\u63db(inverse DFT, IDFT)\uff0c\u5c07\u5176\u8f49\u63db\u6210\u65b0\u7684\u7279\u5fb5\u5e8f\u5217\u3002 \u95dc\u65bc\u4e0a\u8ff0\u4ee5 PLSA \u70ba\u57fa\u790e\u4e4b\u66f4\u65b0\u8abf\u8b8a\u983b\u8b5c\u5f37\u5ea6\u7684\u6f14\u7b97\u6cd5\uff0c\u6709 \u4e0b\u5217\u4e8c\u9805\u5be6\u4f5c\u5c64\u9762\u4e0a\u7684 \u7d30\u7bc0\u9700\u6ce8\u610f\uff0c\u5176\u63cf\u8ff0\u5982\u4e0b\uff1a (1) \u5118\u7ba1\u7279\u5fb5\u6642\u9593\u5e8f\u5217\u4e4b\u9577\u5ea6\u56e0\u8a9e\u53e5\u800c\u7570\uff0c\u4f46\u662f\u5728\u6b64\u6211\u5011\u5c07\u5176\u8abf\u8b8a\u983b\u8b5c\u4e4b\u9577\u5ea6(\u5373\u5176\u53d6\u96e2 \u6563\u5085\u7acb\u8449\u8f49\u63db\u7684\u9ede\u6578)\u8a2d\u70ba\u5b9a\u503c\uff0c\u56e0\u6b64\u6240\u6709\u8a9e\u53e5\u4e4b\u8abf\u8b8a\u983b\u8b5c\u9577\u5ea6\u7686\u76f8\u540c\uff0c\u6b64\u5916\u9700\u6ce8\u610f\u7684 \u662f\uff0c\u6b64\u5b9a\u503c\u9700\u5927\u65bc\u6216\u7b49\u65bc\u4efb\u4e00\u5f85\u8655\u7406\u4e4b\u7279\u5fb5\u6642\u9593\u5e8f\u5217\u7684\u9577\u5ea6\uff0c\u4ee5\u907f\u514d\u6642\u9593\u6df7\u758a(time aliasing)\u7684\u4e0d\u826f\u6548\u61c9\u3002 (2) \u5c0d\u66f4\u65b0\u5f8c\u7684\u8abf\u8b8a\u983b\u8b5c\u9032\u884c\u53cd\u5085\u7acb\u8449\u8f49\u63db\u5f8c\uff0c\u6240\u5f97\u4e4b\u5e8f\u5217\u9577\u5ea6\u6703\u5927\u65bc\u6216\u7b49\u65bc\u539f\u59cb\u7279\u5fb5 \u5e8f\u5217\u7684\u9577\u5ea6(\u5047\u8a2d\u70ba N)\uff0c\u56e0\u6b64\u6211\u5011\u53ea\u4fdd\u7559\u6b64\u65b0\u5e8f\u5217\u7684\u524d N \u9ede\uff0c\u4f5c\u70ba\u6700\u7d42\u7684\u65b0\u7279\u5fb5\u5e8f\u5217\uff0c \u6b64\u4f5c\u6cd5\u662f\u6839\u64da\u6700\u5c0f\u5316\u5e73\u65b9\u5dee(minimum mean squared error, MMSE)\u7684\u6700\u4f73\u6e96\u5247\u800c\u5f97\u3002 \u5728\u5716\u4e09(a)\u8207\u4e09(b)\u4e2d\uff0c\u6211\u5011\u7e6a\u88fd\u4e86\u7531\u4ee5\u4e0a PLSA \u6cd5\u6240\u5f97\u5230\u7684\u4e94\u500b\u96b1\u85cf\u4e3b\u984c\u5c0d\u61c9\u4e4b\u8abf \u8a5e\u53d6\u4ee3\u500b\u6578(substitutions)\u548c\u8a5e\u63d2\u5165\u500b\u6578(insertions)\uff1b\u8a08\u7b97\u516c\u5f0f\u5982\u4e0b\u6240\u793a\uff1a 100% (%) \uf0b4 \uf02d \uf03d \u8f38\u5165\u8a5e\u7e3d\u6578 \u8a5e\u63d2\u5165\u500b\u6578 \u8a5e\u6b63\u78ba\u8fa8\u8b58\u500b\u6578 \u8a5e\u7cbe\u78ba\u7387 (10) \u503c\u5f97\u6ce8\u610f\u7684\u662f\uff0c\u6839\u64da\u539f Aurora-2 \u8cc7\u6599\u5eab\u7684\u8a2d\u5b9a\uff0c\u6bcf\u4e00\u7a2e\u96dc\u8a0a\u7684\u5e73\u5747\u8a5e\u7cbe\u78ba\u7387\u8a08\u7b97\u65b9\u5f0f 2\u3001PLSA \u6cd5\u7d50\u5408\u5176\u4ed6\u5f37\u5065\u6027\u7279\u5fb5\u6f14\u7b97\u6cd5\u6240\u5f97\u4e4b\u8fa8\u8b58\u7387 \u5176\u6b21\uff0c\u6211\u5011\u5c07\u539f\u59cb\u8a9e\u97f3\u7279\u5fb5\u5148\u7d93\u904e\u5012\u983b\u8b5c\u5e73\u5747\u6d88\u53bb\u6cd5(CMS)[1]\u6216\u5012\u983b\u8b5c\u5e73\u5747\u8207\u8b8a\u7570\u6578\u6b63 \u898f\u5316\u6cd5(MVN)[2]\u8655\u7406\u5f8c\uff0c\u518d\u900f\u904e\u6211\u5011\u6240\u63d0\u51fa\u7684 PLSA \u6cd5\u52a0\u4ee5\u8655\u7406\uff0c\u85c9\u6b64\u89c0\u5bdf PLSA \u8207 CMS \u6216 MVN \u9019\u5169\u7a2e\u5178\u578b\u7684\u7279\u5fb5\u5e8f\u5217\u8655\u7406\u6280\u8853\u662f\u5426\u6709\u52a0\u6210\u6027\uff0c\u5176\u6240\u5c0d\u61c9\u7684\u8fa8\u8b58\u7cbe\u78ba\u7387 \u5206\u5225\u5217\u65bc\u8868\u4e8c\u8207\u8868\u4e09\u3002\u89c0\u5bdf\u9019\u5169\u500b\u8868\u7684\u6578\u64da\u3001\u4e26\u8207\u8868\u4e00\u6bd4\u8f03\uff0c\u6211\u5011\u53ef\u77e5\uff1a \u5e73\u5747\u8fa8\u8b58\u7cbe\u78ba\u7387\u8207\u76f8\u5c0d\u932f\u8aa4\u964d\u4f4e\u7387\u3002 \u672c\u8ad6\u6587\u91dd\u5c0d\u8a9e\u97f3\u7279\u5fb5\u6642\u9593\u5e8f\u5217\u4e4b\u8abf\u8b8a\u983b\u8b5c\u63d0\u51fa\u5d84\u65b0\u7684\u5206\u6790\u8207\u5f37\u5316\u6280\u8853\uff0c\u5229\u7528\u6a5f\u7387\u5f0f\u6f5b\u85cf \u8a9e\u610f\u5206\u6790(PLSA)\u8ce6\u4e88\u8abf\u8b8a\u983b\u8b5c\u5f37\u5ea6\u5176\u6a5f\u7387\u7684\u610f\u7fa9\uff0c\u4e26\u900f\u904e\u4e00\u7d44\u6f5b\u85cf\u7684\u4e3b\u984c\u6a5f\u7387\u5206\u5e03\uff0c \u4ee5\u63cf\u7e6a\u8a9e\u53e5\u8207\u8abf\u8b8a\u983b\u8b5c\u5f37\u5ea6\u4e4b\u95dc\u4fc2\uff0c\u540c\u6642\u4e88\u4ee5\u6a5f\u7387\u5f0f\u5206\u89e3\u8207\u6210\u5206\u5206\u6790\uff0c\u4e26\u85c9\u6b64\u66f4\u65b0\u8abf\u8b8a \u983b\u8b5c\u5f37\u5ea6\u4ee5\u6c42\u53d6\u66f4\u5177\u5f37\u5065\u6027\u7684\u8a9e\u97f3\u7279\u5fb5\u5e8f\u5217\u3002\u8fa8\u8b58\u5be6\u9a57\u7d50\u679c\u986f\u793a\uff0c\u6240\u63d0\u51fa\u7684\u65b0\u65b9\u6cd5\u80fd\u6709 \u6548\u63d0\u5347\u96dc\u8a0a\u74b0\u5883\u4e0b\u8a9e\u97f3\u8fa8\u8b58\u7cbe\u78ba\u7387\uff0c\u4e14\u5c55\u793a\u4e86\u6b64\u65b0\u65b9\u6cd5\u8207\u6642\u9593\u5e8f\u5217\u57df\u4e4b\u6b63\u898f\u5316\u6cd5\u80fd\u6709\u4e92 \u8b8a\u983b\u8b5c\u5f37\u5ea6(\u5373\u7b49\u5f0f(4)PLSA \u53ef\u4ee5\u6709\u6548\u5c07\u8a9e\u97f3\u7279\u5fb5\u5e8f\u5217\u91cd\u8981\u7684\u8abf\u8b8a\u983b\u8b5c\u6210\u5206\u64f7\u53d6\u51fa\u3001\u4e26 \u6291\u5236\u4e0d\u91cd\u8981\u6216\u5bb9\u6613\u53d7\u5e72\u64fe\u7684\u4e2d\u9ad8\u983b\u6210\u5206\u3002\u800c\u5716\u4e09(a)\u8207\u4e09(b)\u7684\u4e3b\u8981\u5dee\u5225\uff0c\u5728\u65bc\u5f8c\u8005\u7684\u591a \u6578\u4e3b\u984c\u983b\u8b5c\u5f37\u5ea6\u5176\u6975\u4f4e\u983b\u4e4b\u8fd1\u76f4\u6d41\u6210\u5206(DC)\u5f88\u5c0f\uff0c\u9019\u662f\u56e0\u70ba MVN \u8655\u7406\u5f8c\u7684 MFCC \u7279 \u5fb5\uff0c\u5176\u76f4\u6d41\u6210\u5206\u70ba\u96f6\uff0cPLSA \u6cd5\u6240\u5f97\u7684\u4e3b\u984c\u983b\u8b5c\u5f37\u5ea6\u5fe0\u5be6\u5730\u53cd\u6620\u4e86\u9019\u500b\u524d\u63d0\u3002 \u56db\u3001\u5be6\u9a57\u7d50\u679c\u8207\u5206\u6790 (\u4e00)\u5be6\u9a57\u8a9e\u6599\u5eab \u672c\u8ad6\u6587\u6240\u4f7f\u7528\u7684\u5be6\u9a57\u8a9e\u6599\u5eab\u70ba Aurora-2 \u82f1\u6587\u9023\u7e8c\u6578\u5b57\u8a9e\u6599\u5eab[16]\uff0c\u53c3\u8207\u9304\u97f3\u8a08\u756b\u7684\u8a9e\u8005 \u7686\u662f\u7f8e\u570b\u6210\u5e74\u4eba\u3002\u70ba\u4e86\u8a55\u4f30\u96dc\u8a0a\u6216\u901a\u9053\u5c0d\u65bc\u8a9e\u97f3\u7684\u5f71\u97ff\uff0c\u6e2c\u8a66\u90e8\u5206\u7684\u8a9e\u97f3\u5206\u5225\u647b\u6709\u516b\u7a2e \u4e0d\u540c\u4f86\u6e90\u7684\u52a0\u6210\u6027\u96dc\u8a0a(additive noise)\u548c\u5169\u7a2e\u4e0d\u540c\u7279\u6027\u7684\u901a\u9053\u6548\u61c9\u3002\u6839\u64da\u4e0d\u540c\u7a2e\u985e\u7684\u5e72 \u64fe\uff0c\u5206\u6210\u4e09\u500b\u6e2c\u8a66\u96c6\uff1aSet A, Set B \u8207 Set C\u3002Set A \u7684\u8a9e\u97f3\u5206\u5225\u542b\u6709\u5730\u4e0b\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\uff1b Set 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\u5176\u4e2d\uff0c\u800c\u5176\u4e2d\u7684\u8a0a\u566a\u6bd4\u5247\u6709\u4e03\u7a2e\uff0c\u5206\u5225\u70ba clean ( \uf0a5 dB)\u3001 20 dB\u300115 dB\u300110 dB\u30015 dB\u30010 dB \u548c-5 dB\u3002Aurora-2 \u8cc7\u6599\u5eab\u63d0\u4f9b\u5169\u7a2e\u8a13\u7df4\u8072\u5b78\u6a21\u578b\u7684 \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)\uff0c\u672c\u8ad6\u6587\u7d71\u4e00\u4f7f\u7528\u4e7e\u6de8\u8a9e\u6599\u8a13\u7df4\u6a21\u5f0f\u4f86\u9032\u884c\u5be6\u9a57\uff0c\u8a13\u7df4\u96c6\u7684\u4e7e\u6de8\u8a9e\u97f3\u5171\u6709 8,440 \u53e5\uff0c\u5176\u4e2d\u4e26\u7121\u52a0\u6210\u6027\u96dc\u8a0a\uff0c\u537b\u5305\u542b\u4e86 G.712 \u7684\u901a\u9053\u6548\u61c9\uff0c\u56e0\u6b64\u5728\u4e09\u500b\u6e2c\u8a66\u96c6\u4e2d\uff0c\u8a13\u7df4\u96c6 \u53ea\u8207\u6e2c\u8a66\u96c6 Set C \u6709\u901a\u9053\u4e0a\u7684\u4e0d\u5339\u914d\u3002 (\u4e8c)\u5be6\u9a57\u8a2d\u5b9a \u5728\u524d\u7aef\u8655\u7406\u65b9\u9762\uff0c\u672c\u8ad6\u6587\u7684\u57fa\u790e\u5be6\u9a57\u662f\u63a1\u7528\u6885\u723e\u5012\u983b\u8b5c\u4fc2\u6578\u505a\u70ba\u8a9e\u97f3\u7279\u5fb5\u53c3\u6578\uff0c\u5176\u4e2d\u9810 \u5f37\u8abf(pre-emphasis)\u53c3\u6578\u8a2d\u70ba 0.97\uff0c\u8996\u7a97\u51fd\u6578\u70ba\u6f22\u660e\u7a97(Hamming window)\uff0c\u53d6\u6a23\u97f3\u6846\u9577 \u5ea6(frame length)\u70ba 25 \u6beb\u79d2\uff0c\u97f3\u6846\u9593\u8ddd(frame shift)\u70ba 10 \u6beb\u79d2\uff0c\u6bcf\u500b\u97f3\u6846\u662f\u4ee5 39 \u7dad\u7279\u5fb5 \u5411\u91cf\u8868\u793a\uff0c\u5176\u4e2d\u5305\u542b 12 \u7dad\u7684\u6885\u723e\u5012\u983b\u8b5c\u4fc2\u6578(c 1~c12 )\u8207\u7b2c\u96f6\u7dad\u5012\u983b\u8b5c\u4fc2\u6578(c 0 )\uff0c\u9644\u52a0\u4e0a\u5176 \u7b2c\u4e00\u968e\u5dee\u91cf\u4fc2\u6578(delta coefficient)\u548c\u7b2c\u4e8c\u968e\u5dee\u91cf\u4fc2\u6578(acceleration coefficient)\u3002 \u5728\u8072\u5b78\u6a21\u578b\u7684\u8a2d\u5b9a\u4e0a\uff0c\u6bcf\u500b\u6578\u5b57\u6a21\u578b(one, two, \u2026, nine, zero \u548c oh)\u7686\u7531\u4e00\u500b\u7531\u5de6\u5230 \u53f3(left-to-right)\u5f62\u5f0f\u7684\u9023\u7e8c\u5bc6\u5ea6\u96b1\u85cf\u5f0f\u99ac\u53ef\u592b\u6a21\u578b(continuous density hidden Markov model, CDHMM)\u8868\u793a\uff0c\u5176\u4e2d\u5305\u542b 16 \u500b\u72c0\u614b(state) \uff0c\u6bcf\u500b\u72c0\u614b\u5247\u6709 20 \u500b\u9ad8\u65af\u6df7\u5408(Gaussian mixtures)\u3002\u975c\u97f3\u6a21\u578b\u5247\u70ba 1 \u500b\u72c0\u614b\uff0c\u5167\u542b 36 \u500b\u9ad8\u65af\u6df7\u5408\u7684\u6a21\u578b\u3002\u4e0a\u8ff0\u6240\u6709\u8072\u5b78\u7279\u5fb5\u7684 \u5efa\u7acb\u3001\u8072\u5b78\u6a21\u578b\u7684\u8a13\u7df4\u8207\u5404\u7a2e\u8fa8\u8b58\u5be6\u9a57\u90fd\u662f\u4f7f\u7528 HTK \u5de5\u5177\u5957\u4ef6[18]\u5b8c\u6210\u3002 (\u4e09)\u8fa8\u8b58\u6548\u80fd\u8a55\u4f30\u65b9\u5f0f \u8fa8\uf9fc\u6548\u80fd\u8a55\u4f30\u7684\u65b9\u5f0f\u662f\u63a1\u7528\u7f8e\u570b\u6a19\u6e96\u8207\u79d1\u6280\u7d44\u7e54(the national institute of standards and technology\uff0cNIST)[19]\u6240\u8a02\u7acb\u7684\u8a55\u4f30\u6a19\u6e96\uff0c\u9032\u884c\u6b63\u78ba\u8f49\u8b6f\u6587\u53e5\u5b57\u4e32\u8207\u8fa8\u8b58\u5b57\u4e32\u7684\u6bd4\u8f03\u3002 \u8a55\u4f30\u55ae\u4f4d\u662f\u4ee5\u8a5e\u7cbe\u78ba\u7387(word accuracy)\u70ba\u55ae\u4f4d\uff0c\u8a08\u7b97\u6b63\u78ba\u8f49\u8b6f\u6587\u53e5\u5b57\u4e32\u8207\u8fa8\u8b58\u5b57\u4e32\u9593\u7684 \u662f\u5c0d\u65bc 20 dB \u81f3 0 dB \u7684\u4e94\u7a2e\u8a0a\u566a\u6bd4(SNR)\u8fa8\u8b58\u7387\u53d6\u5e73\u5747\uff0c\u800c\u6392\u9664\u6389\u4e7e\u6de8\u60c5\u6cc1\u548c-5dB \u4e8c \u7a2e\u6975\u7aef\u7684\u8a0a\u566a\u6bd4\u7684\u8fa8\u8b58\u7387\uff1b\u672c\u8ad6\u6587\u5f8c\u7e8c\u7684\u6240\u6709\u5e73\u5747\u8fa8\u8b58\u7387\u7686\u662f\u9075\u5faa\u6b64\u7a2e\u5448\u73fe\u65b9\u5f0f\u3002 (\u56db)\u5be6\u9a57\u7d50\u679c\u5448\u73fe\u8207\u8a0e\u8ad6 1\u3001PLSA \u6cd5\u4f5c\u7528\u65bc\u539f\u59cb MFCC \u6240\u5f97\u4e4b\u8fa8\u8b58\u7387 \u6211\u5011\u5c07\u6240\u63d0\u51fa\u7684 PLSA \u6cd5\u5c0d\u65bc\u539f\u59cb 39 \u7dad MFCC \u8a9e\u97f3\u7279\u5fb5\u53c3\u6578\u6642\u9593\u5e8f\u5217\u505a\u8655\u7406\uff0c\u5176\u5c0d\u61c9 \u7684\u5e73\u5747\u8fa8\u8b58\u7cbe\u78ba\u7387\u8a73\u5217\u65bc\u8868\u4e00\u4e4b\u4e2d\uff1b\u5728 PLSA \u6cd5\u7684\u53c3\u6578\u8a2d\u5b9a\u4e0a\uff0c\u6211\u5011\u4ee4\u6f5b\u85cf\u4e3b\u984c\u500b\u6578 K \u5206\u5225\u70ba 5,10, 15 \u8207 20\uff0c\u800c\u5f0f(9)\u4e2d\u7684\u52a0\u6b0a\u503c \uf061 \u5247\u9810\u8a2d\u70ba 0.85\u3002\u5f9e\u8868\u4e00\u7684\u6578\u64da\u4e2d\uff0c\u6211\u5011\u6709\u4ee5 \u4e0b\u5e7e\u9ede\u767c\u73fe\uff1a (1) \u5728\u5339\u914d\u7684\u4e7e\u6de8\u74b0\u5883\u4e0b\uff0c\u76f8\u8f03\u65bc\u57fa\u790e\u5be6\u9a57\u7d50\u679c\uff0cPLSA \u6cd5\u5c0d\u61c9\u7684\u8fa8\u8b58\u7cbe\u78ba\u7387\u7565\u70ba\u4e0b\u964d\uff0c \u4f46\u5176\u4e0b\u964d\u7a0b\u5ea6\u4e26\u4e0d\u986f\u8457(\u6700\u5927\u4e0b\u964d\u7387\u70ba 0.23%)\uff0c\u4e14\u8ddf\u9078\u64c7\u96b1\u85cf\u4e3b\u984c\u500b\u6578\u4e26\u7121\u660e\u986f\u95dc\u4fc2\u3002 \u6b64\u73fe\u8c61\u9a57\u8b49\u4e86\uff0c\u4ee5 PLSA \u70ba\u57fa\u790e\u7684\u6a21\u578b\u8db3\u4ee5\u5145\u5206\u5448\u73fe\u8a9e\u97f3\u7279\u5fb5\u4e4b\u8abf\u8b8a\u983b\u8b5c\u5f37\u5ea6\u7684\u7279\u6027\u3002 \u5c0d\u65bc\u8abf\u8b8a\u983b\u8b5c\u5f37\u5ea6\u800c\u8a00\uff0cPLSA \u70ba\u4e00\u7a2e\u9ad8\u6548\u80fd\u7de8\u78bc(encoding)\u7684\u65b9\u5f0f\uff0c\u5176\u4e2d\u53ea\u9700\u4f7f\u7528\u5c11\u91cf \u4e4b\u96b1\u85cf\u4e3b\u984c\uff0c\u5c31\u8db3\u4ee5\u4fdd\u6709\u8a9e\u97f3\u7279\u5fb5\u4e4b\u8abf\u8b8a\u983b\u8b5c\u5f37\u5ea6\u5167\u542b\u7684\u8fa8\u8b58\u8cc7\u8a0a\u3002 (2) \u5728\u4e0d\u5339\u914d\u7684\u96dc\u8a0a\u5e72\u64fe\u74b0\u5883\u4e0b\uff0cPLSA \u8655\u7406\u5f8c\u4e4b\u8a9e\u97f3\u7279\u5fb5\u5176\u8868\u73fe\u660e\u986f\u512a\u65bc\u539f\u59cb\u8a9e\u97f3\u7279 \u5fb5\u3002\u8ddf\u57fa\u790e\u5be6\u9a57\u7d50\u679c\u6bd4\u8f03\uff0c\u5728\u4f7f\u7528\u4e3b\u984c\u6578\u70ba 5(K =5)\u7684 PLSA \u6cd5\u6642\uff0c\u8fa8\u8b58\u7cbe\u78ba\u7387\u88ab\u63d0\u5347 \u4e86 17.55%\uff0c\u76f8\u5c0d\u932f\u8aa4\u964d\u4f4e\u7387\u9ad8\u9054 62.84%\uff0c\u800c\u5176\u5b83\u4e3b\u984c\u6578\u7684 PLSA \u6cd5\u4e5f\u6709\u5341\u5206\u985e\u4f3c\u7684\u6548 \u80fd\u3002\u56e0\u6b64\uff0c\u6211\u5011\u6240\u63d0\u51fa\u7684 PLSA \u6cd5\u78ba\u5be6\u80fd\u6709\u6548\u63d0\u5347\u539f\u59cb\u6885\u723e\u5012\u983b\u8b5c\u7279\u5fb5\u4e4b\u96dc\u8a0a\u5f37\u5065\u6027\u3002 \u6b64\u5916\uff0c\u56de\u9867\u5716\u4e09\u6240\u986f\u793a\u4e4b\u4e3b\u984c\u6a5f\u7387\u5206\u5e03\u5c0d\u61c9\u4e4b\u983b\u8b5c\u5f37\u5ea6\uff0c\u53ef\u63a8\u77e5\u85c9\u7531 PLSA \u7684\u8655\u7406\uff0c\u5c0d \u61c9\u81f3\u975e\u8a9e\u97f3\u6210\u5206\u5931\u771f\u4e4b\u9ad8\u983b\u8abf\u8b8a\u983b\u8b5c\u6210\u5206\u6703\u88ab\u5927\u5e45\u7e2e\u6e1b\uff0c\u56e0\u800c\u5f97\u5230\u8fa8\u8b58\u7cbe\u78ba\u7387\u7684\u9032\u6b65\u3002 (3) \u76f8\u8f03\u65bc\u4e7e\u6de8\u74b0\u5883\u60c5\u6cc1\u4e0b\uff0c\u5728\u96dc\u8a0a\u5e72\u64fe\u74b0\u5883\u4e2d\u589e\u52a0\u4e3b\u984c\u6a5f\u7387\u5206\u5e03\u6578\u91cf K\uff0c\u8fa8\u8b58\u7cbe\u78ba\u7387 \u53cd\u800c\u6703\u5fae\u5e45\u4e0b\u964d\u3002\u7136\u800c\u8ddf\u4e7e\u6de8\u74b0\u5883\u4e4b\u60c5\u5f62\u985e\u4f3c\uff0c\u4e0d\u540c\u4e3b\u984c\u6578\u4e4b PLSA \u5176\u6240\u9020\u6210\u7684\u8fa8\u8b58\u7cbe \u78ba\u7387\u5dee\u8ddd\u4e26\u4e0d\u660e\u986f\uff0c\u5e73\u5747\u800c\u8a00\u6700\u5927\u5dee\u8ddd\u50c5 0.30%(\u5f9e 89.62%\u4e0b\u964d\u81f3 89.32%)\u3002 \u5e73\u5747\u8a5e\u7cbe\u78ba\u7387 (%) Clean Set A Set B Set C Avg. RR MFCC baseline 99.79 72.46 68.31 78.82 72.07 \u2500 \u2500 PLSA K=5 99.56 89.20 90.20 89.41 89.62 62.84 K=10 99.59 89.05 90.25 89.25 89.57 62.66 K=15 99.61 88.81 90.15 88.87 89.36 61.90 K=20 99.59 88.78 90.18 88.69 89.32 61.76 Clean Set A Set B Set C Avg. RR 1 RR 2 MFCC baseline 99.79 72.46 68.31 78.82 72.07 --CMS 99.82 79.31 82.46 79.90 80.69 30.86 -CMS K=5 99.66 89.70 91.09 90.00 90.32 65.34 49.93 K=15 99.71 89.49 90.89 89.78 90.11 64.59 48.78 K=20 99.73 89.38 90.92 89.70 90.06 64.41 48.52 \u8868\u4e09\u3001 PLSA \u70ba\u57fa\u790e\u4e4b\u65b9\u6cd5\u4f5c\u7528\u65bc\u7d93 MVN \u8655\u7406\u4e4b MFCC \u7279\u5fb5\u7684\u8fa8\u8b58\u7d50\u679c\uff0c\u5176\u4e2d RR 1 (%)\u8207 RR 2 (%)\u5206\u5225\u70ba\u5c0d\u6bd4\u65bc\u57fa\u790e\u5be6\u9a57\u8207 MVN \u6cd5\u4e4b\u8fa8\u8b58\u7387\u4e4b\u76f8\u5c0d\u932f\u8aa4\u964d\u4f4e\u7387\u3002 \u5e73\u5747\u8a5e\u7cbe\u78ba\u7387 (%) Clean Set A Set B Set C Avg. RR 1 RR 2 MFCC baseline 99.79 72.46 68.31 78.82 72.07 --MVN 99.82 88.58 89.32 88.28 88.82 59.97 -PLSA+ MVN K=5 99.66 89.98 91.01 89.72 90.34 64.70 13.60 K=10 99.68 90.06 91.06 89.90 90.43 65.74 14.40 K=15 99.73 90.07 91.14 89.91 90.47 65.88 14.76 K=20 99.72 90.21 91.18 90.06 90.57 66.24 15.54 3\u3001PLSA \u6cd5\u8207\u5176\u4ed6\u8abf\u8b8a\u983b\u8b5c\u66f4\u65b0\u6cd5\u7684\u6548\u80fd\u6bd4\u8f03 \u5728\u9019\u4e00\u7bc0\u4e2d\uff0c\u6211\u5011\u5c07\u6240\u63d0\u51fa\u7684 PLSA \u6cd5\uff0c\u8207\u4e00\u7cfb\u5217\u7684\u8a9e\u97f3\u7279\u5fb5\u6642\u9593\u5e8f\u5217\u8655\u7406\u6280\u8853\u9032\u884c\u8fa8 \u8b58\u7cbe\u78ba\u7387\u7684\u6bd4\u8f03\u3002\u9019\u4e9b\u6642\u9593\u5e8f\u5217\u8655\u7406\u6280\u8853\uff0c\u5305\u62ec\u4e86\u5728\u7b2c\u4e8c\u7ae0\u4e2d\u63d0\u5230\u7684 HEQ\u3001TSN \u8207 \u4e94\u3001\u7d50\u8ad6\u8207\u672a\u4f86\u5c55\u671b (1) (%) PLSA+ K=10 99.69 89.63 91.00 89.89 90.23 65.02 49.87 \u5716\u56db\u3001PLSA \u6cd5\u8207\u5176\u5b83\u8abf\u8b8a\u983b\u8b5c\u66f4\u65b0\u6cd5\u7684\u6548\u80fd\u6bd4\u8f03(\u7d93 MVN \u524d\u7f6e\u8655\u7406\u904e)</td></tr></table>", | |
| "type_str": "table", | |
| "num": null, | |
| "html": null | |
| } | |
| } | |
| } | |
| } |