| { | |
| "paper_id": "O11-1003", | |
| "header": { | |
| "generated_with": "S2ORC 1.0.0", | |
| "date_generated": "2023-01-19T08:05:33.025853Z" | |
| }, | |
| "title": "", | |
| "authors": [ | |
| { | |
| "first": "Yuan-Fu", | |
| "middle": [], | |
| "last": "\u5ed6\u5143\u752b", | |
| "suffix": "", | |
| "affiliation": { | |
| "laboratory": "", | |
| "institution": "\u570b\u7acb\u53f0\u5317\u79d1\u6280\u5927\u5b78\u96fb\u5b50\u5de5\u7a0b\u7cfb Department of Electronic Engineering National Taipei of Technology \u5433\u8056\u5802 Sheng-Tang Wu", | |
| "location": {} | |
| }, | |
| "email": "" | |
| }, | |
| { | |
| "first": "", | |
| "middle": [], | |
| "last": "Liao", | |
| "suffix": "", | |
| "affiliation": { | |
| "laboratory": "", | |
| "institution": "\u570b\u7acb\u53f0\u5317\u79d1\u6280\u5927\u5b78\u96fb\u5b50\u5de5\u7a0b\u7cfb Department of Electronic Engineering National Taipei of Technology \u5433\u8056\u5802 Sheng-Tang Wu", | |
| "location": {} | |
| }, | |
| "email": "[email protected]" | |
| } | |
| ], | |
| "year": "", | |
| "venue": null, | |
| "identifiers": {}, | |
| "abstract": "", | |
| "pdf_parse": { | |
| "paper_id": "O11-1003", | |
| "_pdf_hash": "", | |
| "abstract": [], | |
| "body_text": [ | |
| { | |
| "text": "EQUATION", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [ | |
| { | |
| "start": 0, | |
| "end": 8, | |
| "text": "EQUATION", | |
| "ref_id": "EQREF", | |
| "raw_str": "\uf028 \uf029 \uf028 \uf029 ct t N s c \uf067 \uf03d \uf0e5 (3) \uf028 \uf029 \uf028 \uf029\uf028 \uf029 c t t c t F s c Y m \uf067 \uf03d\uf02d \uf0e5 (4) \uf028 \uf029 \uf028 \uf029\uf028 \uf029\uf028 \uf029 * c t t c t c t S s diag c Y m Y m \uf067 \uf0e6 \uf0f6 \uf03d \uf02d \uf02d \uf0e7 \uf0f7 \uf0e8 \uf0f8 \uf0e5", | |
| "eq_num": "(5)" | |
| } | |
| ], | |
| "section": "", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "\u5176 \u4f30 \u8a08 \u65b9 \u6cd5 \u76f8 \u540c \uff0c \u4ee5 \u4e0b \u6211 \u5011 \u4ee5 \u5177 \u6709 \u8a9e \u97f3 \u7279 \u6027 \u7684 \u7279 \u5fb5 \u8a9e \u8005 (Eigen-voice)\u6a21\u578b\u70ba\u4f8b\uff0c\u6c42\u53d6 v sp \u3002 \u6211\u5011\u5229\u7528 Expectation Maximization(EM)\u6f14\u7b97\u6cd5\u9032\u884c 10 \u6b21\u7684\u758a\u4ee3\uff0c\u53cd\u8986\u66f4\u65b0 v sp \u4f7f\u4e4b\u8da8\u65bc \u5b9a\u503c\uff1a \u96b1\u85cf\u8b8a\u6578y sp \u4e8b\u5f8c\u5206\u4f48 \u5148\u5047\u8a2d\u96b1\u85cf\u8b8a\u6578y sp (s)\u662f\u5e73\u5747\u503c\u70ba0\u8b8a\u7570\u6578\u70ba1\u7684\u6a19\u6e96\u9ad8\u65af\u5206\u4f48\uff0c\u7576\u6211\u5011\u8f38\u5165\u8a9e\u97f3\u8cc7\u6599 \u5f8c\u5c31\u50cfMAP \u8a9e \u8005 \u8abf \u9069 \u4e00 \u6a23 \u6703 \u6709 \u4e0d \u540c \u7684 \u5206 \u4f48 \u3002 \u6839 \u64da [12] \u5047 \u8a2d \uff0c \u4ee4 \uff0c\u5176\u4e2d \u70bavariance\u4e4b\u8d85\u5411\u91cf\uff0c\u800c\u96b1\u85cf\u8b8a\u6578y sp (s) \u7684\u5206\u4f48\u53ef\u7531\u5e73\u5747\u503c \u8207\u5171\u8b8a\u7570\u6578 \u53bb\u63cf\u8ff0 \u8a72\u6a5f\u7387\u5206\u4f48\u3002 ML\u6cd5\u91cd\u4f30\u8d85\u53c3\u6578 \u521d\u59cb\u53c3\u6578 m \u8207 \u662f\u64f7\u53d6\u81ea average speech model \u7684\u76f8\u95dc\u7d44\u5408\uff0c\u53c3\u6578 v sp \u5247\u662f\u63a1\u7528\u96a8\u6a5f\u7684\u521d \u59cb\u503c\uff0c\u5047\u8a2d\u8a9e\u8005\u70ba s\uff0c\u5b9a\u7fa9\u7d2f\u7a4d\u7684\u7d71\u8a08\u91cf\u5982\u4e0b\uff1a (6) (7) (8) (9) \u5c0d\u6bcf\u4e00\u500b\u6df7\u548c\u6210\u5206 c =1,...,C \u3001\u6bcf\u4e00\u500b\u6df7\u5408\u6210\u5206\u7684\u5143\u7d20 f =1,...,F \uff0c\u8a2d i = ( c -1)F + f\uff0c\u4ee4 u i \u4ee3\u8868 u \u7684\u7b2c i \u5217\uff0c\u800c \u0187 i \u4ee3\u8868 \u0187 \u7684\u7b2c i \u5217\uff0c\u56e0\u6b64\u7279\u5fb5\u7279\u5fb5\u7a7a\u9593 v sp \u7684\u66f4\u65b0\u516c\u5f0f\u53ef\u8868\u793a\u6210\uff1a (i=1,....,", | |
| "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": "\u5c0d\u6bcf\u4e00\u500b\u6df7\u548c\u6210\u5206 c =1,...,C \u3001\u6bcf\u4e00\u500b\u6df7\u5408\u6210\u5206\u7684\u5143\u7d20 f =1,...,F \uff0c\u8a2d i = ( c -1)F + f\uff0c\u4ee4 u i \u4ee3\u8868 u \u7684\u7b2c i \u5217\uff0c\u800c \u0187 i \u4ee3\u8868 \u0187 \u7684\u7b2c i \u5217\uff0c\u56e0\u6b64\u7279\u5fb5\u7279\u5fb5\u7a7a\u9593 d sp \u7684\u66f4\u65b0\u516c\u5f0f\u53ef\u8868\u793a\u6210\uff1a (i=1,....,CF)", | |
| "eq_num": "(15)" | |
| } | |
| ], | |
| "section": "", | |
| "sec_num": null | |
| } | |
| ], | |
| "back_matter": [], | |
| "bib_entries": { | |
| "BIBREF0": { | |
| "ref_id": "b0", | |
| "title": "Frontend Post-Processing and Backend Model Enhancement on the Aurora 2.0/3.0 Databases", | |
| "authors": [ | |
| { | |
| "first": "C", | |
| "middle": [ | |
| "P" | |
| ], | |
| "last": "Chen", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "K", | |
| "middle": [], | |
| "last": "Filali", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "J", | |
| "middle": [], | |
| "last": "Bilmes", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2002, | |
| "venue": "Proc. ICSLP", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "C.P. Chen, K. Filali and J. Bilmes, \"Frontend Post-Processing and Backend Model Enhancement on the Aurora 2.0/3.0 Databases,\" in Proc. ICSLP, 2002.", | |
| "links": null | |
| }, | |
| "BIBREF1": { | |
| "ref_id": "b1", | |
| "title": "Low-resource Noise-Robust Feature Post-Processing on Aurora 2.0", | |
| "authors": [ | |
| { | |
| "first": "C", | |
| "middle": [ | |
| "P" | |
| ], | |
| "last": "Chen", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "J", | |
| "middle": [], | |
| "last": "Bilmes", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "", | |
| "middle": [], | |
| "last": "Kirchhoff", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2002, | |
| "venue": "Proc. ICSLP", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "C.P. Chen, J. Bilmes and K Kirchhoff, \"Low-resource Noise-Robust Feature Post-Processing on Aurora 2.0,\" in Proc. ICSLP, 2002.", | |
| "links": null | |
| }, | |
| "BIBREF2": { | |
| "ref_id": "b2", | |
| "title": "Non-linear transformation of the feature space for robust speech recognition", | |
| "authors": [ | |
| { | |
| "first": "A", | |
| "middle": [], | |
| "last": "De La Torre", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "J", | |
| "middle": [ | |
| "C" | |
| ], | |
| "last": "Segura", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "M", | |
| "middle": [ | |
| "C" | |
| ], | |
| "last": "Benitez", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "A", | |
| "middle": [ | |
| "M" | |
| ], | |
| "last": "Peinado", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "A", | |
| "middle": [ | |
| "J" | |
| ], | |
| "last": "Rubio", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2002, | |
| "venue": "Proc. ICASSP", | |
| "volume": "I", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "A. de la Torre, J. C. Segura, M. C. Benitez, A. M. Peinado and A. J. Rubio, \"Non-linear transformation of the feature space for robust speech recognition,\" in Proc. ICASSP, vol. I, 2002.", | |
| "links": null | |
| }, | |
| "BIBREF3": { | |
| "ref_id": "b3", | |
| "title": "Histogram equalization of speech recognition for robust speech recognition", | |
| "authors": [ | |
| { | |
| "first": "A", | |
| "middle": [], | |
| "last": "De La Torre", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "A", | |
| "middle": [ | |
| "M" | |
| ], | |
| "last": "Peinado", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "J", | |
| "middle": [ | |
| "C" | |
| ], | |
| "last": "Segura", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "J", | |
| "middle": [ | |
| "L P" | |
| ], | |
| "last": "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": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "A. de la Torre, A. M. Peinado, J. C. Segura, J. L. P. Cordoba, M. C. Benitez and A. J. Rubio, \"Histogram equalization of speech recognition for robust speech recognition,\" IEEE Trans. on Speech and Audio Processing, vol. 13, no. 3, 2005.", | |
| "links": null | |
| }, | |
| "BIBREF4": { | |
| "ref_id": "b4", | |
| "title": "Speech processing, transmission and quality aspects", | |
| "authors": [ | |
| { | |
| "first": "", | |
| "middle": [], | |
| "last": "Etsi", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "", | |
| "middle": [], | |
| "last": "Document", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": null, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "ETSI standard document, \"Speech processing, transmission and quality aspects (STQ);", | |
| "links": null | |
| }, | |
| "BIBREF5": { | |
| "ref_id": "b5", | |
| "title": "distributed speech recognition; extended advanced front-end feature extraction algorithm", | |
| "authors": [], | |
| "year": null, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "distributed speech recognition; extended advanced front-end feature extraction algorithm;", | |
| "links": null | |
| }, | |
| "BIBREF6": { | |
| "ref_id": "b6", | |
| "title": "compression algorithm; back-end reconstruction algorithm", | |
| "authors": [], | |
| "year": 2003, | |
| "venue": "ETSI Standard ES", | |
| "volume": "202", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "compression algorithm; back-end reconstruction algorithm,\" ETSI Standard ES 202 212, 2003.", | |
| "links": null | |
| }, | |
| "BIBREF7": { | |
| "ref_id": "b7", | |
| "title": "Maximum likelihood linear regression for speaker adaptation of continuous density hidden Markov models", | |
| "authors": [ | |
| { | |
| "first": "C", | |
| "middle": [ | |
| "J" | |
| ], | |
| "last": "Leggetter", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "P", | |
| "middle": [ | |
| "C" | |
| ], | |
| "last": "Woodland", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1995, | |
| "venue": "Computer Speech Lang", | |
| "volume": "9", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "C.J. Leggetter and P.C. Woodland, \"Maximum likelihood linear regression for speaker adaptation of continuous density hidden Markov models,\" Computer Speech Lang., vol. 9, 1995.", | |
| "links": null | |
| }, | |
| "BIBREF8": { | |
| "ref_id": "b8", | |
| "title": "Maximum a Posteriori estimation for multivariate Gaussian mixture observations of Markov chains", | |
| "authors": [ | |
| { | |
| "first": "J", | |
| "middle": [ | |
| "L" | |
| ], | |
| "last": "Gauvain", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "C", | |
| "middle": [ | |
| "H" | |
| ], | |
| "last": "Lee", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1994, | |
| "venue": "IEEE Trans.on Speech Audio Processing", | |
| "volume": "2", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "J.L. Gauvain and C.H. Lee, \"Maximum a Posteriori estimation for multivariate Gaussian mixture observations of Markov chains,\" IEEE Trans.on Speech Audio Processing, vol. 2, 1994.", | |
| "links": null | |
| }, | |
| "BIBREF9": { | |
| "ref_id": "b9", | |
| "title": "Robust continuous speech recognition using parallel model combination", | |
| "authors": [ | |
| { | |
| "first": "M", | |
| "middle": [], | |
| "last": "Gales", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "S", | |
| "middle": [], | |
| "last": "Young", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1996, | |
| "venue": "IEEE Trans. on Speech and Audio Processing", | |
| "volume": "13", | |
| "issue": "3", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "M. Gales and S. Young, \"Robust continuous speech recognition using parallel model combination,\" IEEE Trans. on Speech and Audio Processing, vol. 13, no. 3, September 1996.", | |
| "links": null | |
| }, | |
| "BIBREF10": { | |
| "ref_id": "b10", | |
| "title": "Large-Vocabulary Speech Recognition under Adverse Acoustic Environments", | |
| "authors": [ | |
| { | |
| "first": "L", | |
| "middle": [], | |
| "last": "Deng", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "A", | |
| "middle": [], | |
| "last": "Acero", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "M", | |
| "middle": [], | |
| "last": "Plumpe", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "X", | |
| "middle": [], | |
| "last": "Huang", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2000, | |
| "venue": "Proc. ICSLP", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "L. Deng, A. Acero, M. Plumpe and X. Huang. \"Large-Vocabulary Speech Recognition under Adverse Acoustic Environments,\" in Proc. ICSLP 2000.", | |
| "links": null | |
| }, | |
| "BIBREF11": { | |
| "ref_id": "b11", | |
| "title": "Mean and variance adaptation within the MLLR framework", | |
| "authors": [ | |
| { | |
| "first": "M", | |
| "middle": [ | |
| "J F" | |
| ], | |
| "last": "Gales", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "P", | |
| "middle": [ | |
| "C" | |
| ], | |
| "last": "Woodland", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1996, | |
| "venue": "Computer Speech Lang", | |
| "volume": "10", | |
| "issue": "3", | |
| "pages": "249--264", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "M.J.F. Gales and P.C. Woodland, \"Mean and variance adaptation within the MLLR framework,\" Computer Speech Lang., vol. 10, no. 3, pp. 249-264, 1996.", | |
| "links": null | |
| }, | |
| "BIBREF12": { | |
| "ref_id": "b12", | |
| "title": "Joint Factor Analysis of Speaker and Session Variability : Theory and AlgorithmsMontreal", | |
| "authors": [ | |
| { | |
| "first": "P", | |
| "middle": [], | |
| "last": "Kenny", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2005, | |
| "venue": "CRIM", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "P. Kenny, \"Joint Factor Analysis of Speaker and Session Variability : Theory and AlgorithmsMontreal\", Technical report CRIM-06/08-13 Montreal, CRIM, 2005", | |
| "links": null | |
| }, | |
| "BIBREF13": { | |
| "ref_id": "b13", | |
| "title": "A Study of Inter-Speaker Variability in Speaker Verification", | |
| "authors": [ | |
| { | |
| "first": "P", | |
| "middle": [], | |
| "last": "Kenny", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "P", | |
| "middle": [], | |
| "last": "Ouellet", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "N", | |
| "middle": [], | |
| "last": "Dehak", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "V", | |
| "middle": [], | |
| "last": "Gupta", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "P", | |
| "middle": [], | |
| "last": "Dumouchel", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2008, | |
| "venue": "IEEE Transactions on Audio Speech and Language Processing", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Kenny, P., Ouellet, P., Dehak, N., Gupta, V., and Dumouchel, P., \"A Study of Inter-Speaker Variability in Speaker Verification,\" IEEE Transactions on Audio Speech and Language Processing, July 2008.", | |
| "links": null | |
| }, | |
| "BIBREF14": { | |
| "ref_id": "b14", | |
| "title": "A maximization technique occurring in the statistical analysis of probabilistic functions of Markov chains", | |
| "authors": [ | |
| { | |
| "first": "L", | |
| "middle": [ | |
| "E" | |
| ], | |
| "last": "Baum", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "T", | |
| "middle": [], | |
| "last": "Petrie", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "G", | |
| "middle": [], | |
| "last": "Soules", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "N", | |
| "middle": [], | |
| "last": "Weiss", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1970, | |
| "venue": "Ann. Math. Statist", | |
| "volume": "41", | |
| "issue": "1", | |
| "pages": "164--171", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "L. E. Baum, T. Petrie, G. Soules, and N. Weiss, \"A maximization technique occurring in the statistical analysis of probabilistic functions of Markov chains\", Ann. Math. Statist., vol. 41, no. 1, pp. 164-171, 1970.", | |
| "links": null | |
| } | |
| }, | |
| "ref_entries": { | |
| "TABREF0": { | |
| "html": null, | |
| "text": "\u6458\u8981 \u6458\u8981-\u672c\u8ad6\u6587\u4e3b\u8981\u7814\u7a76\u65bc\u5f37\u5065\u6027\u8a9e\u97f3\u8fa8\u8a8d\u4e0a\uff0c\u6211\u5011\u63d0\u51fa\u806f\u5408\u8a9e\u8005\u3001\u96dc\u8a0a\u74b0\u5883\u8207\u8a9e\u97f3\u5167\u5bb9 \u56e0\u7d20\u5206\u6790(Joint Speaker and Noisy Environment and Speech Content Factor Analysis\uff1bJSEC)\uff0c \u4e3b\u8981\u662f\u900f\u904e\u806f\u5408\u56e0\u7d20\u5206\u6790\uff0c\u5728\u7279\u5fb5\u7a7a\u9593\u505a\u5373\u6642\u8a9e\u97f3\u8fa8\u8a8d\u6a21\u578b\u88dc\u511f(online recognition model compensation)\uff0c\u4f7f\u5f97\u8abf\u9069\u51fa\u4f86\u7684\u6a21\u578b\u8207\u6e2c\u8a66\u74b0\u5883\u80fd\u5920\u76e1\u91cf\u5339\u914d\uff0c\u9032\u800c\u63d0\u5347\u8fa8\u8b58\u6548 \u679c\u3002\u6b64\u5916\uff0c\u6211\u5011\u5148\u5c07 JSEC \u5206\u89e3\u6210\u8a9e\u97f3\u548c\u975e\u8a9e\u97f3\u4e8c\u500b\u6a21\u578b\u505a\u6a21\u578b\u8abf\u9069\u3001\u4f30\u7b97\u5f71\u97ff\u56e0\u7d20\uff0c \u63a5\u8457\u6bcf\u500b\u6a21\u578b\u518d\u5229\u7528\u968e\u5c64\u5f0f\u7684\u6982\u5ff5\uff0c\u8a9e\u97f3\u7279\u6027\u8003\u616e\u4e4b\u56e0\u7d20\u5206\u6210\u96dc\u8a0a\u74b0\u5883\u7279\u5fb5\u7a7a\u9593\u3001\u8a9e\u8005 \u7279\u5fb5\u7a7a\u9593\u3001\u8aaa\u8a71\u5167\u5bb9\u7279\u5fb5\u7a7a\u9593\u8207\u7368\u7279\u56e0\u7d20\u7a7a\u9593\u5206\u5225\u4f30\u7b97\uff0c\u975e\u8a9e\u97f3\u7279\u6027\u8003\u616e\u4e4b\u56e0\u7d20\u5247\u5206\u6210 \u96dc\u8a0a\u7279\u5fb5\u7a7a\u9593\u548c\u7368\u7279\u56e0\u7d20\u7a7a\u9593\u5206\u5225\u4f30\u7b97\uff0c\u6700\u5f8c\u518d\u628a\u8a9e\u97f3\u548c\u975e\u8a9e\u97f3\u7d44\u5408\u56de\u8fa8\u8a8d\u7528\u7684\u6a21\u578b\uff0c \u7528\u6b64\u65b9\u5f0f\u4f86\u964d\u4f4e\u6211\u5011\u7684\u53c3\u6578\u91cf\u3002\u6211\u5011\u4f7f\u7528 Aurora2 \u8a9e\u6599\u5eab\u505a\u5be6\u9a57\uff0c\u5728\u8907\u5408\u60c5\u5883\u7684\u8a13\u7df4\u6a21 \u5f0f\u4e0b\uff0c\u6211\u5011\u5f97\u5230\u6700\u4f73\u7684\u8fa8\u8b58\u932f\u8aa4\u7387\u70ba 4.37%\uff0c\u6bd4\u50b3\u7d71\u5f37\u5065\u6027\u53c3\u6578\u6c42\u53d6\u65b9\u6cd5 MVA (Mean subtraction\uff0cVariance normalization\uff0cand ARMA filtering)[1][2]\u7684\u932f\u8aa4\u7387 4.99%\u4f4e\u4e86\u8a31\u591a\uff0c \u4e5f\u6bd4\u6211\u5011\u5148\u524d\u63d0\u51fa\u7684 JSE (Joint Speaker and Noisy Environment Factor Analysis)[11]\u65b9\u6cd5 \u7684\u932f\u8aa4\u7387\u76f8\u7576\u751a\u81f3\u597d\u4e00\u9ede\u3002\u9664\u4e86\u8fa8\u8a8d\u7387\u4e4b\u5916\uff0c\u6211\u5011\u63d0\u51fa\u7684\u65b9\u6cd5\u4e5f\u80fd\u4f7f\u5f97\u8abf\u9069\u6a21\u578b\u7684\u53c3\u6578 \u91cf\u5927\u5e45\u4e0b\u964d\uff0cJSEC \u53c3\u6578\u91cf\u7d04\u70ba\u50b3\u7d71 MVA \u7684 4 \u500d\uff0c\u4e5f\u6bd4 JSE \u65b9\u6cd5\u5c11\u4e86\u5341\u5206\u4e4b\u4e00\u7684\u53c3\u6578 \u91cf\uff0c\u56e0\u6b64\u70ba\u66f4\u6709\u6548\u7387\u7684\u8abf\u9069\u65b9\u6cd5\u3002 \u5176\u4e2d\uff1a m sp \u3001m non \uff1a\u7531\u8a9e\u97f3\u53c3\u6578\u4e32\u63a5\u800c\u6210\u7684\u8d85\u5411\u91cf\uff0c\u6a21\u578b\u53c3\u6578\u4e32\u63a5\u800c\u6210\u7684\u8d85\u5411\u91cf\u3002 x sp \u3001x non \uff1a\u7279\u5fb5\u96dc\u8a0a\u74b0\u5883\u7a7a\u9593\u7684\u6295\u5f71\u91cf\uff0c\u521d\u59cb\u5047\u8a2d\u5e73\u5747\u503c\u70ba 0 \u8b8a\u7570\u6578\u70ba 1 \u7684\u6a19\u6e96\u9ad8\u65af\u5206\u4f48\u3002 y sp \uff1a\u7279\u5fb5\u8a9e\u8005\u7279\u5fb5\u7a7a\u9593\u7684\u6295\u5f71\u91cf\uff0c\u521d\u59cb\u5047\u8a2d\u5e73\u5747\u503c\u70ba 0 \u8b8a\u7570\u6578\u70ba 1 \u7684\u6a19\u6e96\u9ad8\u65af\u5206\u4f48\u3002 r sp \uff1a\u8a9e\u97f3\u5167\u5bb9\u7279\u5fb5\u7a7a\u9593\u7684\u6295\u5f71\u91cf\uff0c\u521d\u59cb\u5047\u8a2d\u5e73\u5747\u503c\u70ba 0 \u8b8a\u7570\u6578\u70ba 1 \u7684\u6a19\u6e96\u9ad8\u65af\u5206\u4f48\u3002 z sp \u3001z non \uff1a\u7368\u7279\u56e0\u7d20\u7279\u5fb5\u7a7a\u9593\u7684\u6295\u5f71\u91cf\uff0c\u521d\u59cb\u5047\u8a2d\u5e73\u5747\u503c\u70ba 0 \u8b8a\u7570\u6578\u70ba 1 \u7684\u6a19\u6e96\u9ad8\u65af\u5206\u4f48\u3002 u sp \u3001u non \uff1a\u7279\u5fb5\u96dc\u8a0a\u74b0\u5883\u7279\u5fb5\u7a7a\u9593\u3002 \u3001v sp \u3001g sp \u3001d sp \u8868\u793a\u3002\u975e\u5177\u6709\u8a9e\u97f3\u7279\u6027\u7684\u7247\u6bb5\u8a9e\u53e5\uff0c\u5247\u8a13\u7df4\u4e00\u500b non speech \u7684\u6a21\u578b\uff0c \u4e26\u4e14\u50c5\u5c0d\u4e0d\u540c\u96dc\u8a0a\u505a\u6a19\u8a18\uff0c\u540c\u65bc\u5177\u6709\u8a9e\u97f3\u7279\u6027\u7684\u90e8\u5206\u3002\u63a5\u8457\u4f9d\u5e8f\u4f30\u7b97\u96dc\u8a0a\u8207\u7368\u7279\u56e0\u7d20\u7279 \u5fb5\u7a7a\u9593\uff0c\u5206\u5225\u4ee5 u non \u8207 d non \u8868\u793a\u3002 \u5716\u4e8c\u3001JSEC \u4e4b\u7cfb\u7d71\u6d41\u7a0b\u5716 \u5728\u6e2c\u8a66\u7aef\uff0c\u6211\u5011\u4f30\u7b97\u6e2c\u8a66\u8a9e\u6599\u5177\u6709\u8a9e\u97f3\u7279\u6027\u5f71\u97ff\u56e0\u7d20\u7684\u6295\u5f71\u91cf x sp \u3001y sp \u3001z sp \uff0c\u7136\u5f8c\u6295\u5f71 \u5230\u5efa\u7acb\u597d\u7684 u sp \u3001v sp \u3001d sp \uff0c\u5f97\u5230\u504f\u79fb\u91cf u sp x sp + v sp y sp + d sp z sp \uff1b\u8207\u975e\u5177\u6709\u8a9e\u97f3\u7279\u6027\u5f71\u97ff\u56e0 \u7d20\u7684\u6295\u5f71\u91cf x non \u3001z non \uff0c\u7136\u5f8c\u6295\u5f71\u5230\u5efa\u7acb\u597d\u7684 u non \u3001d non \uff0c\u5f97\u5230\u504f\u79fb\u91cf u non x non + d non z non \u3002 \u5f97\u5230\u5169\u8005\u504f\u79fb\u91cf\u5f8c\uff0c\u53e6\u5916\u518d\u4f30\u7b97\u8a13\u7df4\u7aef\u5177\u6709\u8a9e\u97f3\u7279\u6027\u4e4b\u8aaa\u8a71\u5167\u5bb9\u7684\u504f\u79fb\u91cf g sp r sp \uff0c\u7528\u610f", | |
| "num": null, | |
| "content": "<table><tr><td>\u4e00\u3001\u7dd2\u8ad6</td><td/></tr><tr><td colspan=\"2\">\u8a9e\u97f3\u8fa8\u8a8d\u7cfb\u7d71\u53d7\u96dc\u8a0a\u74b0\u5883\u3001\u8a9e\u8005\u7279\u6027\u8207\u901a\u9053\u6548\u61c9\u7b49\u5f71\u97ff\uff0c\u5c0e\u81f4\u8fa8\u8b58\u7387\u4e0b\u964d\u3002\u901a\u5e38\u8655\u7406\u9019</td></tr><tr><td colspan=\"2\">\u4e9b\u5f71\u97ff\u56e0\u7d20\u6216\u74b0\u5883\u4e0d\u5339\u914d\u554f\u984c\uff0c\u6709\u5169\u7a2e\u8f03\u5e38\u898b\u7684\u65b9\u6cd5\uff1a\u5f37\u5065\u6027\u8a9e\u97f3\u53c3\u6578\u6c42\u53d6(robust</td></tr><tr><td colspan=\"2\">speech feature extraction)\u8207\u8a9e\u97f3\u6a21\u578b\u8abf\u9069(speech model adaptation)\u3002</td></tr><tr><td colspan=\"2\">\u5f37 \u5065 \u6027 \u53c3 \u6578 \u6c42 \u53d6 \u4e4b \u65b9 \u6cd5 \uff0c \u6211 \u5011 \u53ef \u4ee5 \u8209 \u5e7e \u500b \u7d93 \u5178 \u7684 \u4f8b \u5b50 \uff1a \u5012 \u983b \u8b5c \u6b63 \u898f \u5316 ARMA</td></tr><tr><td colspan=\"2\">(Auto-regression and Moving Average)\u6ffe\u6ce2\u6280\u8853(Mean subtraction\uff0cVariance normalization\uff0c and ARMA filtering\uff1bMVA)[1][2]\u3001\u5206\u5e03\u7b49\u5316\u6cd5(Histogram Equalization\uff1bHEQ)[3][4]\uff0c\u8207 \u5169\u968e\u5f0f\u7dad\u7d0d\u6ffe\u6ce2\u5668(two-stage Wiener filter)[5]\u7b49\uff0c\u5b83\u5011\u7684\u7279\u9ede\u662f\u6709\u6548\u4e14\u5bb9\u6613\u5be6\u73fe\u3002 \u81f3\u65bc\u6a21\u578b\u8abf\u9069\u65b9\u6cd5\uff0c\u53c8\u53ef\u5206\u70ba\u662f\u5426\u9700\u8981\u5148\u9a57\u77e5\u8b58\uff0c\u4e0d\u9700\u8981\u5148\u9a57\u77e5\u8b58\u7684\u65b9\u6cd5\uff0c\u4e3b\u8981\u6709\uff1a\u6700 \u5927\u76f8\u4f3c\u5ea6\u7dda\u6027\u56de\u6b78(Maximum Likelihood Linear Regression\uff1bMLLR)[6]\u3001\u6700\u5927\u4e8b\u5f8c\u6a5f\u7387\u8abf \u9069\u6cd5(Maximum A Posteriori \uff1bMAP)[7]\u8abf\u9069\u6cd5\u3001\u5e73\u884c\u6a21\u578b\u7d50\u5408(Parallel Model Combination\uff1b PMC)[8]\u7b49\uff0c\u4ee5\u4e0a\u7686\u70ba\u7d93\u5178\u4e14\u5e38\u898b\u4e4b\u8a9e\u97f3\u6a21\u578b\u8abf\u9069\u6cd5\uff0c\u7d93\u5e38\u88ab\u61c9\u7528\u65bc\u8a9e\u97f3\u548c\u8a9e\u8005\u8fa8\u8a8d\u7cfb \u7d71\u3002 \u800c\u9700\u5148 \u9a57\u77e5 \u8b58\u7684\u65b9 \u6cd5\uff0c \u5e38\u898b\u7684\u65b9 \u6cd5 \u5982 \u96d9\u8072 \u6e90\u70ba \u57fa\u790e\u4e4b \u5206\u6bb5 \u7dda\u6027 \u88dc\u511f (Stereo-based Piecewise Linear Compensation, SPLICE)[9]\u3002\u6211\u5011\u4e5f\u66fe\u5229\u7528\u4e8b\u5148\u6536\u96c6\u5927\u91cf\u8a9e\u8005\u8207\u74b0\u5883\u5148 \u9a57\u77e5\u8b58\uff0c\u63d0\u51fa\u57fa\u65bc\u96dc\u8a0a\u74b0\u5883\u53c3\u8003\u6a21\u578b\u5167\u63d2\u6cd5(Reference Model Weighting\uff1bRMW)\u3001\u96dc\u8a0a \u7279\u5fb5\u6700\u5927\u76f8\u4f3c\u5ea6\u7dda\u6027\u8ff4\u6b78(Eigen-Maximum Likelihood Linear Regression\uff1bEMLLR)[10]\uff0c \u57fa\u65bc\u806f\u5408\u8a9e\u8005\u8207\u96dc\u8a0a\u74b0\u5883\u56e0\u7d20\u5206\u6790 (Joint Speaker and Noisy Environment Factor Analysis\uff1b \u5716\u4e00\u3001JSEC \u67b6\u69cb\u5716 v sp \uff1a\u7279\u5fb5\u8a9e\u8005\u7279\u5fb5\u7a7a\u9593\u3002 d sp \u3001d non \uff1a\u7368\u7279\u56e0\u7d20\u7279\u5fb5\u7a7a\u9593\u3002 g sp \uff1a\u8a9e\u97f3\u5167\u5bb9\u7279\u5fb5\u7a7a\u9593\u3002 2.2 JSEC \u7cfb\u7d71\u67b6\u69cb \u5716\u4e8c\u662f JSEC \u4e4b\u7cfb\u7d71\u6d41\u7a0b\u5716\uff0c\u5728\u8a13\u7df4\u7aef\uff0c\u6211\u5011\u5c07\u8a13\u7df4\u8a9e\u6599\u505a Force-alignment\uff0c\u8b8a\u6210\u4e0d\u540c \u8a9e\u97f3\u5167\u5bb9\u7684\u7247\u6bb5\u8a9e\u53e5\u3002\u5177\u6709\u8a9e\u97f3\u7279\u6027\u7684\u7247\u6bb5\u8a9e\u53e5\u8a13\u7df4\u4e00\u500b\u540d\u70ba speech \u7684\u8072\u5b78\u6a21\u578b\uff0c\u6211 \u5011\u4fbf\u662f\u5229\u7528\u9019\u7a2e\u65b9\u5f0f\u4f86\u964d\u4f4e\u6700\u5f8c\u91cd\u5efa\u6a21\u578b\u4e4b\u53c3\u6578\u91cf\u3002\u4e26\u4e14\u5c0d\u5177\u6709\u8a9e\u97f3\u7279\u6027\u7684\u7247\u6bb5\u8a9e\u53e5\u505a \u6a19\u8a18\u5206\u985e\uff0c\u63a5\u8457\u518d\u4f9d\u5e8f\u4f30\u7b97\u96dc\u8a0a\u3001\u8a9e\u8005\u3001\u8aaa\u8a71\u5167\u5bb9\uff0c\u6700\u5f8c\u662f\u7368\u7279\u56e0\u7d20\u7684\u7279\u5fb5\u7a7a\u9593\uff0c\u5206\u5225 \u6a21\u578b\uff0c\u7136\u5f8c\u518d\u52a0\u4e0a\u4ee5\u672a\u5207\u5272\u8a9e\u6599\u8a13\u7df4\u7684\u8072\u5b78\u6a21\u578b\u3001\u975c\u97f3\u6a21\u578b\u8207\u505c\u9813\u6a21\u578b\u90e8\u5206\uff0c\u5373\u53ef\u91cd\u5efa \u51fa\u6bcf\u4e00\u53e5\u6e2c\u8a66\u8a9e\u6599\u7368\u6709\u7684\u6a21\u578b\uff0c\u6700\u5f8c\u505a\u8fa8\u8b58\u7d50\u679c\u3002 \u5728\u5f97\u5230\u6240\u9700\u8981\u7684\u8f49\u63db\u77e9\u9663\u4e4b\u5f8c\u5c31\u53ef\u4ee5\u9032\u884c\u7b2c\u4e8c\u6b65\u9a5f\uff0c\u4e5f\u5c31\u662f\u5c07\u539f\u59cb\u53c3\u6578\u5411\u91cf\u5c0d\u8f49\u63db\u77e9\u9663 \u505a\u5167\u7a4d\u904b\u7b97\uff0c\u5c31\u80fd\u8f49\u63db\u6210\u65b0\u53c3\u6578\u5411\u91cf\uff0c\u7136\u5f8c\u9001\u9032\u6a21\u578b\u8a13\u7df4\u3002\u63a5\u4e0b\u4f86\u5169\u6bb5\u5c07\u6558\u8ff0\u4e3b\u6210\u5206\u5206 \u6790\u548c\u7dda\u6027\u9451\u5225\u5206\u6790\u7684\u539f\u7406\u548c\u5be6\u4f5c\u7684\u6b65\u9a5f\u3002 2.3 \u7279\u5fb5\u7a7a\u9593\u7684\u4f30\u8a08 \u6211\u5011\u985e\u4f3c\u65bc\u53c3\u8003\u6587\u737b[12]\u4e4b\u53e4\u5178 MAP\u3001\u7279\u5fb5\u8a9e\u8005\u548c\u7279\u5fb5\u901a\u9053\u7b49\u65b9\u6cd5\uff0c\u8868\u793a\u5404\u7a2e\u56e0\u7d20\u7684\u95dc \u4fc2\u3002\u800c\u7531\u4e0d\u540c\u9ad8\u65af\u6df7\u5408\u5206\u5e03(mixture)\u7684\u5171\u8b8a\u7570\u6578\u4e32\u63a5\u800c\u6210\u7684\u5c0d\u89d2\u77e9\u9663\u5247\u53ef\u4f5c\u70ba\u53c3\u6578\u4f30\u6e2c \u4ee5 u sp \u662f\u628a\u55ae\u4e00\u7684\u8072\u5b78\u6a21\u578b\uff0c\u53ef\u4ee5\u4f9d\u7167\u4e0d\u540c\u8aaa\u8a71\u5167\u5bb9\u4e4b\u5f71\u97ff\uff0c\u8abf\u9069\u70ba\u4e0d\u540c\u8a9e\u97f3\u5167\u5bb9\u7279\u6027\u7684\u8072\u5b78 \u7684\u521d\u59cb\u503c\u3002</td></tr><tr><td colspan=\"2\">JSE)[11]\u7b49\u65b9\u6cd5\uff0c\u6548\u679c\u7686\u76f8\u7576\u4e0d\u932f\u3002 JSEC \u5728\u8a13\u7df4\u7aef\u5c07\u8a13\u7df4\u8a9e\u6599\u5206\u6210\u5169\u985e\uff0c\u4e00\u985e\u70ba\u5de6\u908a\u5177\u6709\u8a9e\u97f3\u7279\u6027\u4e4b\u8a9e\u53e5\u505a\u5f71\u97ff\u56e0\u7d20\u4e4b\u5206</td></tr><tr><td colspan=\"2\">\u985e\uff0c\u53e6\u4e00\u985e\u70ba\u53f3\u908a\u975e\u8a9e\u97f3\u7279\u6027\u4e4b\u8a9e\u53e5\u505a\u5f71\u97ff\u56e0\u7d20\u4e4b\u5206\u985e\uff0c\u518d\u5229\u7528\u968e\u5c64\u5f0f\u7684\u6982\u5ff5\uff0c\u5c07\u8a9e\u97f3 \u672c\u6587\u6240\u63d0\u5230\u7684\u806f\u5408\u56e0\u7d20\u5206\u6790\u662f\u53c3\u8003[13]\u7684\u4f5c\u6cd5\uff0c\u5c07\u8a9e\u97f3\u6a21\u578b\uff0c\u5229\u7528\u64f7\u53d6 average speech \u6b64\u8ad6\u6587\u6211\u5011\u63d0\u51fa\u4e86\u9700\u5148\u9a57\u77e5\u8b58\u7684\u8a9e\u8005\u3001\u96dc\u8a0a\u74b0\u5883\u8207\u8a9e\u97f3\u5167\u5bb9\u56e0\u7d20\u5206\u6790\u4e4b\u5f37\u5065\u6027\u8a9e\u97f3\u8fa8\u8a8d \u7279\u6027\u5206\u6210\u4e86\u96dc\u8a0a\u74b0\u5883\u7279\u5fb5\u7a7a\u9593\u3001\u8a9e\u8005\u7279\u5fb5\u7a7a\u9593\u3001\u8aaa\u8a71\u5167\u5bb9\u7279\u5fb5\u7a7a\u9593\u8207\u7368\u7279\u56e0\u7d20\u7a7a\u9593\u5206\u5225 model \u7684\u5e73\u5747\u503c\u6240\u69cb\u6210\u7684\u8d85\u5411\u91cf\u4f5c\u70ba\u57fa\u6e96\uff0c\u5c31\u50cf\u662f\u50b3\u7d71\u7684 MAP \u8a9e\u8005\u8abf\u9069\u4e00\u6a23\uff0c\u800c\u7531\u4e0d (Joint Speaker and Noisy Environment and Speech Content Factor Analysis\uff1bJSEC)\uff0c JSEC \u4f30\u7b97\uff0c\u975e\u8a9e\u97f3\u7279\u6027\u70ba\u96dc\u8a0a\u7279\u5fb5\u7a7a\u9593\u8207\u7368\u7279\u56e0\u7d20\u7a7a\u9593\u5206\u5225\u4f30\u7b97\uff0c\u6700\u5f8c\u518d\u628a\u8a9e\u97f3\u548c\u975e\u8a9e\u97f3\u7d44 \u540c\u6df7\u5408\u6210\u5206\u7684\u5171\u8b8a\u7570\u6578 \u4e32\u63a5\u800c\u6210\u7684\u5c0d\u89d2\u77e9\u9663 \u5247\u53ef\u4f5c\u70ba\u53c3\u6578\u4f30\u6e2c\u7684\u521d\u59cb\u503c\u3002\u5728\u6a21\u578b\u53c3 \u4e3b\u8981\u904b\u7528\u5728\u96dc\u8a0a\u5206\u6790\u8655\u7406\uff0c\u6240\u8003\u616e\u7684\u5f71\u97ff\u56e0\u7d20\u53ca\u4f30\u7b97\u9806\u5e8f\u5982\u5716\u4e00\u7684 JSEC \u67b6\u69cb\u5716\u3002 \u5408\u56de\u8fa8\u8a8d\u7528\u7684\u6a21\u578b\u3002\u7576\u6211\u5011\u5f97\u5230\u4e0d\u540c\u5f71\u97ff\u56e0\u7d20\u7684\u7a7a\u9593\u5f8c\uff0c\u6700\u5f8c\u5728\u6e2c\u8a66\u7aef\uff0c\u8f38\u5165\u6e2c\u8a66\u8a9e\u6599 \u6578\u4f30\u6e2c\u4e4b\u524d\u5148\u5b9a\u7fa9\u7cfb\u7d71\u7684\u6a5f\u7387\u5047\u8a2d\u3002</td></tr><tr><td colspan=\"2\">\u5f8c\uff0c\u6e2c\u8a66\u8a9e\u6599\u5c0d\u500b\u5225\u7279\u5fb5\u7a7a\u9593\u505a\u6295\u5f71\uff0c\u5373\u53ef\u5c0d\u6a21\u578b\u505a\u5373\u6642\u7684\u8abf\u9069\u3002</td></tr><tr><td>\u6ce2\u6c0f\u7d71\u8a08</td><td/></tr><tr><td colspan=\"2\">\u4e8c\u3001\u806f\u5408\u8a9e\u8005\u3001\u96dc\u8a0a\u74b0\u5883\u8207\u8a9e\u97f3\u5167\u5bb9\u56e0\u7d20\u5206\u6790 \u6211\u5011\u5148\u4f7f\u7528\u6ce2\u6c0f\u7d71\u8a08[14]\u4e3b\u8981\u662f\u4ee5average speech model\u7684\u5e73\u5747\u503c\u3001\u8b8a\u7570\u6578\u4ee5\u53ca\u6b0a\u91cd\u4f86\u8a08</td></tr><tr><td>\u7b97\u6a5f\u7387\u7d71\u8a08\u91cf\u3002\u5047\u8a2d\u8a9e\u8005s\u4ee5\u53ca\u8a9e\u8005\u7279\u5fb5\u5411\u91cf</td><td>\uff0c\u5c0d\u65bc\u6bcf\u4e00\u500b\u6df7\u5408\u6210\u5206c\uff0c\u6211\u5011</td></tr><tr><td>2.1 JSEC \u6a21\u578b\u8868\u793a \u5b9a\u7fa9\u6ce2\u6c0f\u7d71\u8a08\u5982\u4e0b\uff1a</td><td/></tr><tr><td colspan=\"2\">JSEC \u4e3b\u8981\u8003\u616e\u6e2c\u8a66\u8a9e\u6599\u5728\u8fa8\u8b58\u6642\uff0c\u53d7\u5230\u96dc\u8a0a\u74b0\u5883\u3001\u8a9e\u8005\u3001\u8a9e\u97f3\u5167\u5bb9\u8207\u5176\u4ed6\u56e0\u7d20\u7684\u5f71\u97ff\u3002</td></tr><tr><td colspan=\"2\">\u800c JSEC \u6bd4 JSE \u591a\u8003\u616e\u7684\u8a9e\u97f3\u5167\u5bb9\u5f71\u97ff\uff0c\u53ef\u5206\u70ba\u5177\u6709\u8a9e\u97f3\u7279\u6027\u7684\u90e8\u5206\uff0c\u4ee5\u53ca\u975e\u5177\u6709\u8a9e\u97f3</td></tr><tr><td>\u7279\u6027\u3002</td><td/></tr><tr><td colspan=\"2\">\u6211\u5011\u5b9a\u7fa9\u5177\u6709\u8a9e\u97f3\u7279\u6027\u4e4b JSEC \u6a21\u578b\u662f\u7531\u53e4\u5178 MAP(Classical MAP)\u3001\u7279\u5fb5\u96dc\u8a0a\u74b0\u5883\u3001\u7279</td></tr><tr><td colspan=\"2\">\u5fb5\u8a9e\u8005\u3001\u8a9e\u97f3\u5167\u5bb9(zero~nine\u3001oh\u3001silence)\u56db\u500b\u6a21\u578b\u7d50\u5408\u800c\u6210\uff1a</td></tr><tr><td>M speech = m sp + u sp x sp + v sp y sp + g sp r sp + d sp z sp</td><td>(1)</td></tr><tr><td colspan=\"2\">\u800c\u975e\u8a9e\u97f3\u7279\u6027\u4e4b JSEC \u6a21\u578b\u662f\u7531\u7279\u5fb5\u96dc\u8a0a\u8207\u53e4\u5178 MAP(Classical MAP)\u6a21\u578b\u7d50\u5408\u800c\u6210\uff1a</td></tr><tr><td>\u95dc\u9375\u8a5e\uff1a\u5f37\u5065\u6027\u8a9e\u97f3\u8fa8\u8a8d\uff0c\u56e0\u7d20\u5206\u6790\uff0cAurora2 M nonspeech = m non +u non x non + d non z non</td><td>(2)</td></tr></table>", | |
| "type_str": "table" | |
| }, | |
| "TABREF1": { | |
| "html": null, | |
| "text": "(c=1,...,C)\u6240\u69cb\u6210\u3002\u8a2dF(s)\u70baCF x 1\u7684\u8d85\u5411\u91cf\uff0c\u662f\u7531\u6bcf\u4e00\u500b (s) (c=1,...,C)\u4e32\u63a5\u800c\u6210\u3002 \u8a2dS(s)\u70baCF x CF\u7684\u5c0d\u89d2\u77e9\u9663\uff0c\u5176\u4e2d\u7684\u5c0d\u89d2\u5340\u584a\u662f\u7531 (s) (c=1,...,C)\u6240\u69cb\u6210\u3002 2.3.1 \u8a9e\u8005\u3001\u96dc\u8a0a\u74b0\u5883\u3001\u8a9e\u97f3\u5167\u5bb9\u7279\u5fb5\u7a7a\u9593 \u6c42\u5f97\u6ce2\u5f0f\u7d71\u8a08\u91cf\u4e4b\u5f8c\uff0c\u7531\u53c3\u8003\u6587\u737b[13]\u6211\u5011\u53ef\u4ee5\u8a08\u7b97\u51fa\u5177\u6709\u8a9e\u97f3\u7279\u6027\u7684\u8a9e\u8005\u3001\u96dc\u8a0a\u74b0\u5883 \u7279\u5fb5\u7a7a\u9593\uff0c\u548c\u975e\u8a9e\u97f3\u7279\u6027\u7684\u96dc\u8a0a\u74b0\u5883\u7279\u5fb5\u7a7a\u9593\u3002 \u8a9e\u8005\u3001\u96dc\u8a0a\u74b0\u5883\u3001\u8a9e\u97f3\u5167\u5bb9\u7279\u5fb5\u7a7a\u9593\u4f30\u7b97\u65b9\u6cd5\u76f8\u540c\uff0c\u4f46\u662f\u8a9e\u97f3\u5167\u5bb9\u7b97\u51fa\u4f86\u7684\u96b1\u85cf\u8b8a\u6578 r sp \uff0c\u5fc5\u9808\u5132\u5b58\u8d77\u4f86\u7d66\u6e2c\u8a66\u7aef\u4f7f\u7528\uff0c\u56e0\u70ba\u5728\u8fa8\u8a8d\u7684\u6642\u5019\u4e26\u4e0d\u77e5\u9053\u8981\u8aaa\u54ea\u4e9b\u8a9e\u97f3\u5167\u5bb9,\u5148\u5047 \u8a2d\u6bcf\u4e00\u500b model \u5e73\u5747\u503c\u5728\u54ea\u88e1\uff0c\u518d\u5229\u7528 ML \u6cd5\u91cd\u4f30\u8d85\u53c3\u6578\u53d6\u5f97 g sp \u4e4b\u5f8c\uff0c\u5373\u53ef\u5c07\u8aaa\u8a71\u5167 \u5bb9\u6295\u5f71\u5230\u5c0d\u61c9\u4f4d\u7f6e\u3002 \u7531\u65bc u sp , v sp , g sp , u non", | |
| "num": null, | |
| "content": "<table><tr><td>(s)</td></tr><tr><td>\u5176\u4e2d\uff1a</td></tr><tr><td>\u4ee3\u8868\u8a9e\u8005\u7279\u5fb5\u5411\u91cf\u65bc\u6642\u9593\u6642\u843d\u65bc\u6df7\u5408\u6210\u5206\u7684\u4e8b\u5f8c\u6a5f\u7387\uff0c\u800c \u4ee3\u8868average speech</td></tr><tr><td>model\u4e2d\u6df7\u5408\u6210\u5206\u7684c\u5e73\u5747\u503c\u3002\u63a5\u8457\u8a2dN(s)\u70baCF x CF\u7684\u5c0d\u89d2\u77e9\u9663\uff0c\u5176\u4e2d\u7684\u5c0d\u89d2\u5340\u584a\u662f\u7531</td></tr></table>", | |
| "type_str": "table" | |
| }, | |
| "TABREF3": { | |
| "html": null, | |
| "text": "\u4e0a\u8ff0\u7684\u8868\u793a\u5f0f\uff0c\u53ef\u4ee5\u5f9e\u53c3\u8003\u6587\u737b[13]\u5f97\u5230\u76f8\u95dc\u7684\u8868\u793a\u3002 2.3.3 \u6295\u5f71\u91cf x, y, r, z \u7684\u4f30\u8a08 \u7576\u6211\u5011\u5728\u8a13\u7df4\u7aef\u5f97\u5230\u6c42\u53d6\u51fa\u7684\u5177\u6709\u8a9e\u97f3\u7279\u6027\u7684\u8d85\u53c3\u6578 u sp \u3001v sp \u3001g sp \u3001d sp \uff0c\u4ee5\u53ca\u975e\u8a9e\u97f3\u7279 \u6027\u7684\u8d85\u53c3\u6578 u non \u3001d non \u5f8c\uff0c\u6e2c\u8a66\u7aef\u7684\u53c3\u6578\u518d\u4f9d\u7167\u5177\u6709\u8a9e\u97f3\u7279\u6027\u4e4b\u5f71\u97ff\u56e0\u7d20\uff0c\u7d93\u904e\u4f30\u7b97\u800c \u5f97\u5230\u500b\u5225\u96dc\u8a0a\u5f71\u97ff\u4e4b\u6295\u5f71\u91cf x sp \u3001\u8a9e\u8005\u5f71\u97ff\u4e4b\u6295\u5f71\u91cf y sp \u3001\u8aaa\u8a71\u5167\u5bb9\u4e4b\u6295\u5f71\u91cf r sp \u3001\u7368\u7279\u56e0 \u7d20\u4e4b\u6295\u5f71\u91cf z sp \uff0c\u975e\u8a9e\u97f3\u7279\u6027\u4e4b\u5f71\u97ff\u56e0\u7d20\u4e00\u6a23\u7d93\u904e\u4f30\u7b97\u800c\u5f97\u5230\u6295\u5f71\u91cf x non \u3001z non \u3002 \u5f97\u5230 x sp \u3001y sp \u3001r sp \u3001z sp \u5f8c\uff0c\u6295\u5f71\u5230 u sp \u3001v sp \u3001g sp \u3001d sp \u7279\u5fb5\u7a7a\u9593\uff0c\u5373\u53ef\u5c0d\u6a21\u578b\u505a\u5373\u6642\u7684\u8abf \u9069\uff0c\u91cd\u5efa\u51fa\u6bcf\u53e5\u6e2c\u8a66\u8a9e\u6599\u7368\u6709\u7684\u8fa8\u8a8d\u6a21\u578b\uff0c\u800c\u8b8a\u7570\u6578\u3001\u8f49\u79fb\u6a5f\u7387\u8207\u6b0a\u91cd\u4e4b\u5f71\u97ff\u5f88\u5c0f\uff0c\u6545 \u66ab\u4e14\u5047\u8a2d\u4e0d\u8003\u616e\u8b8a\u7570\u6578\u3001\u8f49\u79fb\u6a5f\u7387\u8207\u6b0a\u91cd\u7b49\u554f\u984c\u3002 \u4e09\u3001\u5be6\u9a57\u7d50\u679c\u8207\u5206\u6790 \u6211\u5011\u53e6\u5916\u505a\u4e86\u4e00\u7d44 complex backend \u5be6\u9a57\uff0c\u628a mixture \u6578\u5f9e 3 \u62c9\u5230 20\uff0c\u6211\u5011\u4e00\u6a23\u4f7f\u7528\u6700 \u4f73\u7684\u96dc\u8a0a 20 \u7dad\u3001\u8a9e\u8005 60 \u7dad\u3001\u8a9e\u97f3\u5167\u5bb9 8 \u7dad\uff0c\u6700\u5f8c\u5be6\u9a57\u6578\u64da\u5982\u5716\u5341\u516d\u548c\u5716\u5341\u4e03\u3002 \u5716\u5341\u516b\u3001complex backend MVA \u8207\u6539\u8b8a\u53c3\u6578\u91cf\u7684 JSEC \u6bd4\u4f8b\u5716 \u5f9e\u5716\u5341\u516b\u6211\u5011\u53ef\u4ee5\u89c0\u5bdf\u51fa\u53c3\u6578\u91cf\u4f9d\u7136\u9060\u5c0f\u65bc\u5148\u524d\u7684 JSE\uff0cJSE \u6bd4 JSEC \u591a 9-10 \u500d\u7684\u53c3 \u6578\u91cf\uff0c\u800c JSEC \u53ea\u6bd4 MVA \u591a\u4e86 4 \u500d\u7684\u53c3\u6578\u91cf\uff0c\u904b\u7b97\u91cf\u4f4e\u5f88\u591a\u3002 \u56db\u3001\u7d50\u8ad6 \u672c\u8ad6\u6587\u7684\u4e3b\u8981\u7814\u7a76\u76ee\u6a19\u662f\u63d0\u51fa\u65b0\u7684 JSEC \u65b9\u6cd5\uff0c\u4e26\u4e14\u4f7f\u7528 Aurora2 \u505a\u5be6\u9a57\uff0c\u5be6\u9a57\u6578\u64da\u6700 \u5f8c\u505a\u4e86\u7e3d\u6574\u7406\u5982\u5716\u5341\u4e5d\u548c\u5716\u4e8c\u5341\u3002\u7531\u5716\u4e2d\u6211\u5011\u63d0\u51fa\u7684 JSEC \u5728 complex backend \u7684\u5be6\u9a57 \u7576\u4e2d\uff0c\u6211\u5011\u767c\u73fe\u53ef\u4ee5\u548c\u539f\u4f86 JSE \u7cfb\u7d71\u7684\u932f\u8aa4\u7387\u5dee\u4e0d\u591a\uff0c\u751a\u81f3\u53ef\u4ee5\u66f4\u4f4e\u4e00\u4e9b\uff0c\u800c\u4e14\u4e5f\u6bd4\u50b3 \u7d71\u65b9\u6cd5 MVA \u932f\u8aa4\u7387 4.99%\u4f4e\u4e86\u8a31\u591a\uff1b\u9664\u4e86\u8fa8\u8a8d\u7387\u4e4b\u5916\uff0c\u6211\u5011\u63d0\u51fa\u7684\u65b9\u6cd5\u4e5f\u80fd\u4f7f\u8abf\u9069\u6a21 \u578b\u7684\u53c3\u6578\u91cf\u6bd4\u539f\u4f86\u7684 JSE \u964d\u4f4e\u4e86\u5341\u5206\u4e4b\u4e00\uff0cJSE \u7684\u53c3\u6578\u91cf\u662f MVA \u7684 40 \u500d\uff0c\u4f46 JSEC \u53ea \u6bd4 MVA \u591a\u4e86 4 \u500d\u7684\u53c3\u6578\u91cf\u800c\u5df2\uff0c\u56e0\u6b64\u70ba\u66f4\u6709\u6548\u7387\u7684\u8abf\u9069\u65b9\u6cd5\u3002.", | |
| "num": null, | |
| "content": "<table><tr><td>\u8a9e\u97f3\u7279\u6027\u548c\u975e\u8a9e\u97f3\u7279\u6027\u4e4b\u6295\u5f71\u91cf\u7b97\u6cd5\u4e00\u6a23\uff0c\u6211\u5011\u4ee5\u4f30\u7b97\u8a9e\u97f3\u90e8\u5206\u7684\u6295\u5f71\u91cf\u70ba\u4f8b\uff1a \u96dc\u8a0a\u5f71\u97ff\u4e4b\u6295\u5f71\u91cfx sp \u5047\u8a2d x sp = E[x(s)] = \u8a9e\u8005\u5f71\u97ff\u4e4b\u6295\u5f71\u91cfy sp \u5047\u8a2d y sp = E[y(s)] = \u8a9e\u97f3\u5167\u5bb9\u5f71\u97ff\u4e4b\u6295\u5f71\u91cfr sp \u5047\u8a2d r sp = E[r (s)] = \u7368\u7279\u56e0\u7d20\u4e4b\u6295\u5f71\u91cfz sp \u5047\u8a2d \u53e6\u5916\u6211\u5011\u5be6\u9a57\u5c0d\u7167\u9700\u8981\u7528\u5230\u6211\u5011\u5148\u524d\u63d0\u51fa\u7684 JSE \u65b9\u6cd5\uff0c\u5176\u6a21\u578b\u53ef\u8868\u793a\u70ba\uff1a M = m + ux(s) + vy(s) + dz(s) 6.30% 6.35% 4.35% \u6700\u5f8c\u6211\u5011\u53c3\u6578\u8a2d\u5b9a\u4f7f\u7528\u6700\u4f73\u7684\u96dc\u8a0a 20 \u7dad\u3001\u8a9e\u8005 60 \u7dad\u3001\u8a9e\u97f3\u5167\u5bb9 8 \u7dad\uff0c\u548c\u5176\u4ed6\u7cfb\u7d71\u65b9\u6cd5 (16) (17) (18) (19) (20) (21) (24) 15.00% MVA 4.40% 4.45% 4.50% \u932f\u8aa4\u7387 3.4 \u5be6\u9a57\u7d50\u679c\u8207\u8a0e\u8ad6 \u800c JSEC \u53ea\u6bd4 MVA \u591a\u4e86 4 \u500d\u7684\u53c3\u6578\u91cf\uff0c\u6548\u80fd\u66f4\u597d\u3001\u904b\u7b97\u91cf\u66f4\u5c0f\u3002 20.00% JSEC \u8a9e\u8005\u6700\u4f73\u7dad\u5ea6 z sp = E[z(s)] = (23) \u8868\u4e00\u3001\u8907\u5408\u60c5\u5883\u8a13\u7df4\u6a21\u5f0f\u5404\u7a2e\u53c3\u6578\u7d44\u5408\u8fa8\u8b58\u7d50\u679c 4.55% 4.60% 4.65% \u7531\u5716\u5341\u4e8c\uff0c\u6211\u5011\u53ef\u4ee5\u770b\u5230\u8a9e\u97f3\u5167\u5bb9\u7684\u6700\u4f73\u7dad\u5ea6\u662f 8 \u7dad\u3002 \u578b\u6240\u9700\u7684\u53c3\u6578\u91cf\u660e\u986f\u6bd4\u539f\u672c JSE \u7684\u65b9\u6cd5\u964d\u4f4e\u5f88\u591a\uff0cJSE \u6bd4 JSEC \u591a 9 -10 \u500d\u7684\u53c3\u6578\u91cf\uff0c \u5716\u4e8c\u5341\u3001\u53c3\u6578\u91cf\u6bd4\u8f03\u4e4b\u6bd4\u4f8b\u5716 \u5716\u5341\u516d\u3001complex backend \u5404\u7cfb\u7d71\u65b9\u6cd5\u4e0d\u540c\u74b0\u5883\u4e4b\u6bd4\u8f03\u5716 \u4f46\u5728\u8abf\u9069\u6a21\u578b\u6240\u9700\u7684\u53c3\u6578\u91cf\u7684\u65b9\u9762\uff0c\u5982\u8868\u4e8c\u548c\u5716\u5341\u4e94\u3002\u6211\u5011\u53ef\u4ee5\u5f9e\u5716\u5341\u4e94\u767c\u73fe\uff0c\u8abf\u9069\u6a21 (22) 3.1 \u5be6\u9a57\u8a2d\u5b9a \u672c\u8ad6\u6587\u5be6\u9a57\u662f\u4ee5\u570b\u969b\u4e0a\u5ee3\u6cdb\u7528\u5728\u96dc\u8a0a\u74b0\u5883\u8a9e\u97f3\u8fa8\u8b58\u6280\u8853\u5f37\u5065\u6027\u7684\u6a19\u6e96\u8a9e\u6599\u5eab Aurora 2 \u70ba\u4e3b\u3002Aurora2 \u662f\u4ee5 TIDigits \u70ba\u57fa\u790e\uff0c\u52a0\u4e0a\u4e0d\u540c\u96dc\u8a0a\u4ee5\u53ca\u901a\u904e\u7279\u5b9a\u7684\u901a\u9053\u6548\u61c9\u88fd\u6210\u3002 Aurora2 \u662f\u4e00\u500b\u9023\u7e8c\u6578\u5b57\u4e32\u8a9e\u6599\u5eab\uff0c\u6bcf\u53e5\u97f3\u6bb5\u5305\u542b\u4e00\u81f3\u4e03\u500b\u9023\u7e8c\u6578\u5b57\uff0c\u9577\u5ea6\u6700\u591a\u4e0d\u8d85\u904e \u4e09\u79d2\u9418\u3002 \u8a9e\u6599\u9996\u5148\u901a\u904e\u7406\u60f3\u6ffe\u6ce2\u5668\u5c07 20 kHz \u964d\u983b\u70ba 8 kHz\uff0c\u6b64\u70ba\u5b9a\u7fa9\u7684\u4e7e\u6de8(Clean)\u8a9e\u6599\uff0c\u6bcf\u500b\u4e7e \u6de8\u97f3\u6bb5\u5148\u7d93\u7279\u5b9a\u7684\u901a\u9053\u6548\u61c9\uff0c\u518d\u4f9d\u5404\u7a2e\u8a0a\u96dc\u6bd4(SNR20\u3001SNR15\u3001SNR 10\u3001SNR 5\u3001SNR 0 \u548c SNR -5dB)\u52a0\u4e0a\u4e0d\u540c\u7684\u52a0\u6210\u6027\u96dc\u8a0a\u3002 \u8a13\u7df4\u8a9e\u6599\u6df7\u5408\u5404\u7a2e\u8a0a\u96dc\u6bd4\u53ca\u4e0d\u540c\u74b0\u5883\u96dc\u8a0a\u7684\u8907\u5408\u60c5\u5883\u8a13\u7df4\u8a13\u7df4\u6a21\u5f0f\u3002\u6e2c\u8a66\u8a9e\u6599\u90e8\u5206\u5247 \u662f\u4f9d\u7167\u539f\u672c Aurora2 \u81ea\u884c\u5efa\u7acb\u7684\u4e0d\u540c\u901a\u9053\u6548\u61c9\u8207\u52a0\u6210\u6027\u96dc\u8a0a\uff0c\u5171\u5206\u6210 A\u3001B\u3001C \u4e09\u7a2e \u6e2c\u8a66\u7d44\u5408\u3002 \u672c\u8ad6\u6587\u63a1\u7528\u6885\u723e\u5012\u983b\u8b5c\u4fc2\u6578\uff0c\u53ca\u8072\u5b78\u6a21\u578b\u70ba\u9023\u7e8c\u6027\u5bc6\u5ea6\u7684\u96b1\u85cf\u5f0f\u99ac\u53ef\u592b\u6a21\u578b\uff0c\u6a21\u578b \u7684\u72c0\u614b\u8f49\u79fb\u53ea\u505c\u7559\u5728\u539f\u59cb\u72c0\u614b\uff0c\u53ca\u7531\u5de6\u81f3\u53f3\u8f49\u79fb\u5230\u4e0b\u4e00\u500b\u76f8\u9130\u7684\u72c0\u614b\u3002 \u6578\u5b57\u8072\u5b78\u6a21\u578b\u7684\u55ae\u4f4d\u70ba\u5168\u8a5e\u6a21\u578b\uff0c\u5341\u4e00\u500b\u82f1\u6587\u6578\u5b57\u8072\u5b78\u6a21\u578b(0\uff5e9 \u548c oh)\u3002\u6bcf\u500b\u8072\u5b78\u6a21 \u578b\u6709 16 \u500b\u72c0\u614b\uff0c\u6bcf\u500b\u72c0\u614b\u542b 3 \u500b\u9ad8\u65af\u5206\u5e03\u6a21\u578b\u3002\u9664\u6578\u5b57\u8072\u5b78\u6a21\u578b\u5916\uff0c\u9084\u6709\u975c\u97f3(silence) \u6a21\u578b\u548c\u505c\u9813(short pause)\u6a21\u578b\u3002\u8fa8\u8b58\u6548\u80fd\u8a55\u4f30\u4e0a\uff0c\u63a1\u53d6\u8fa8\u8b58\u8a5e\u932f\u8aa4\u7387\uff0c\u9019\u8003\u616e\u4e86\u522a\u9664\u578b \u932f\u8aa4\u3001\u63d2\u5165\u578b\u932f\u8aa4\u548c\u53d6\u4ee3\u578b\u932f\u8aa4\u3002\u6211\u5011\u5be6\u9a57\u5206\u70ba\u4e8c\u985e\uff1asimple backend (3 mixture)\u548c complex backend (20 mixture)\uff0c\u5982\u8868\u4e00\u6240\u793a\u3002 Backend Speech model Silence model Short pause model Simple 16 state, each state 3 mixture 3 state, each state 6 mixture 1 state, each state 6 mixture Complex 16 state, each state 20 mixture 3 state, each state 64 mixture 1 state, each state 64 mixture backend \u5206\u6790\u7279\u5fb5\u7a7a\u9593\u4e26\u756b\u51fa\u7279\u5fb5\u7a7a\u9593\u6295\u5f71\u5716\uff0c\u76ee\u7684\u662f\u8b93\u5404\u7a2e\u4e0d\u540c\u96dc\u8a0a\u7684\u6e2c\u8a66\u8a9e\u6599\uff0c\u80fd \u6295\u5f71\u5230\u6b63\u78ba\u7684\u7279\u5fb5\u7a7a\u9593\u4e0a\u3002\u5728\u5efa\u69cb\u7279\u5fb5\u7a7a\u9593\u6642\uff0c\u6211\u5011\u5148\u4f30\u7b97\u5df2\u7d93\u505a\u597d\u5206\u985e\u8cc7\u8a0a\u7684\u7d71\u8a08\u91cf\uff0c \u63a5\u8457\u518d\u4f9d\u7167 u\u3001v\u3001g\u3001d \u4e4b\u9806\u5e8f\uff0c\u9010\u4e00\u4f30\u7b97\u500b\u5225\u4e4b\u7279\u5fb5\u7a7a\u9593\u3002\u7232\u4e86\u65b9\u4fbf\u5206\u6790\uff0c\u6211\u5011\u53d6\u524d \u5169\u7dad\u7684\u7279\u5fb5\u5411\u91cf\u4f5c x \u8ef8\u548c y \u8ef8\uff0c\u5efa\u69cb\u4e00\u500b\u4e8c\u7dad\u7a7a\u9593\uff0c\u9996\u5148\u4ee5\u96dc\u8a0a\u985e\u578b(\u5730\u4e0b\u9435\u3001\u4eba\u8072\u3001 \u6c7d\u8eca\u3001\u5c55\u89bd\u6703)\u505a\u5206\u6790\uff0c\u6211\u5011\u63a1\u7528\u4e03\u7a2e SNR(clean\u3001SNR20\u3001SNR15\u3001SNR10\u3001SNR5\u3001 SNR0\u3001SNR-5)\u505a\u7279\u5fb5\u7a7a\u9593\u5206\u6790\uff0c\u5176\u7d50\u679c\u5982\u5716\u4e09\u3001\u5716\u56db\u6240\u793a\u3002 \u5716\u4e09\u3001simple backend JSEC \u8a9e\u97f3\u7279\u6027\u4e4b\u96dc\u8a0a\u7279\u5fb5\u7a7a\u9593\u6295\u5f71\u5716 \u8d8a\u4f86\u8d8a\u660e\u986f\uff0c\u800c\u672b\u7aef\u7684\u7dda\u689d\u4fbf\u8ddf\u8457\u9010\u6f38\u5206\u958b\u3002\u7531\u4ee5\u4e0a\u4e4b\u7279\u5fb5\u7a7a\u9593\u6295\u5f71\u5716\uff0c\u6211\u5011\u5f97\u77e5\u6c42\u51fa \u4f86\u4e4b\u7279\u5fb5\u7a7a\u9593\u80fd\u5920\u6709\u6548\u5730\u5c07\u9019\u4e9b\u5e72\u64fe\u56e0\u7d20\u500b\u5225\u5206\u958b\uff0c\u63d0\u5347\u8fa8\u8b58\u6548\u80fd\u3002 \u63a5\u8457\u6211\u5011\u8981\u505a\u8a9e\u8005\u7279\u6027\u5206\u6790\uff0c\u5982\u5716\u4e94\uff1a \u5716\u4e94\u3001simple backend JSEC \u8a9e\u8005\u7279\u5fb5\u7a7a\u9593\u6295\u5f71\u5716 \u5728\u5716\u4e94\u4e2d\uff0c\u6211\u5011\u53ef\u4ee5\u770b\u5230\u5176\u6295\u5f71\u7d50\u679c\uff0c\u5f88\u660e\u986f\u4f9d\u7167\u8a9e\u8005\u7684\u4e0d\u540c\u88ab\u5206\u6210\u5169\u908a\uff0c\u6211\u5011\u4ee5\u300co\u300d \u8207\u300c+\u300d\u7684\u7b26\u865f\u5206\u5225\u8868\u793a\u7537\u751f\u8207\u5973\u751f\u7684\u7279\u6027\u3002 \u6700\u5f8c\u662f\u8a9e\u97f3\u5167\u5bb9\u4e4b\u7279\u5fb5\u7a7a\u9593\u6295\u5f71\u5206\u6790\uff0c\u5176\u7d50\u679c\u5982\u5716\u516d\u6240\u793a\uff1a \u5716\u516d\u3001simple backend JSEC \u8a9e\u97f3\u5167\u5bb9\u4e4b\u7279\u5fb5\u7a7a\u9593\u6295\u5f71\u5716 \u6211\u5011\u53ef\u4ee5\u770b\u5230\u5716\u516d\u5176\u6295\u5f71\u7d50\u679c\uff0c\u4f9d\u7167\u8a9e\u97f3\u5167\u5bb9\u88ab\u5206\u958b\uff0c\u800c\u6bd4\u8f03\u985e\u4f3c\u7684\u97f3\uff0c\u4f8b\u5982 oh\u3001four\uff0c \u4f3c\u4e4e\u6703\u6bd4\u8f03\u9760\u8fd1\uff0c\u800c\u8907\u5408\u60c5\u5883\u7684\u9ede(digit)\u5927\u7d04\u662f\u5728\u6240\u6709\u9ede\u7684\u5e73\u5747\u4f4d\u7f6e\u3002\u7531\u4ee5\u4e0a\u4e09\u7a2e\u4e0d\u540c \u5f71\u97ff\u56e0\u7d20\u4e4b\u7279\u5fb5\u7a7a\u9593\u6295\u5f71\u5716\uff0c\u6211\u5011\u9810\u6e2c\u8fa8\u8b58\u6548\u679c\u61c9\u7576\u4e0d\u932f\u3002 3.3 simple backend \u6211\u5011\u7684\u5be6\u9a57\u70ba\u4e86\u8981\u6709\u6548\u7387\u7684\u627e\u51fa\u7279\u5fb5\u7a7a\u9593\u7684\u6700\u4f73\u7dad\u5ea6\uff0c\u9996\u5148\u56fa\u5b9a\u8a9e\u8005(S) 55 \u7dad\uff0c\u8a9e\u97f3\u5167 \u5bb9(T) 6 \u7dad\u3002\u96dc\u8a0a(N)\u7dad\u5ea6\u5171 40 \u7dad\uff0c\u6240\u4ee5\u6211\u5011\u5f9e 20 \u7dad\u958b\u59cb\u627e\u6700\u4f73\u6548\u679c\uff0c\u4e26\u4e14\u4e00\u6b21\u5f80\u4e0a\u6216 \u5f80\u4e0b\u589e\u52a0 6 \u7dad(14 \u7dad\u300120 \u7dad\u548c 24 \u7dad)\u5c0b\u627e\u6700\u4f73\u7dad\u5ea6\u3002\u53e6\u5916\uff0c\u7531\u65bc\u8abf\u9069\u6a21\u578b\u4e2d\u7684\u8b8a\u7570\u6578\u3001 \u8f49\u79fb\u6a5f\u7387\u8207\u6b0a\u91cd\u5f71\u97ff\u5f88\u5c0f\uff0c\u56e0\u6b64\u5148\u5047\u8a2d\u8207\u6bd4\u8f03\u7684 MVA\u3001JSE \u76f8\u540c\uff0c\u4e26\u4e14\u628a\u5be6\u9a57\u5206\u6210 simple \u5716\u516b\u3001simple backend JSEC \u8a9e\u8005\u6700\u4f73\u7dad\u5ea6\u6bd4\u8f03\u5716 \u5716\u516b\u6211\u5011\u53ef\u4ee5\u770b\u5230\u8a9e\u8005\u7684\u6700\u4f73\u7dad\u5ea6\u662f 60 \u7dad\uff0c\u56e0\u6b64\u6211\u5011\u63a5\u8457\u56fa\u5b9a\u96dc\u8a0a 20 \u7dad\uff0c\u8a9e\u8005\u53d6 60 \u7dad\uff0c\u518d\u53d6\u8a9e\u97f3\u5167\u5bb9 6 \u7dad\u30018 \u7dad\u548c 10 \u7dad\uff0c\u505a\u6e2c\u8a66\u53ef\u4ee5\u5f97\u5230\u4ee5\u4e0b\u7d50\u679c\uff1a 6.21% JSEC \u8a9e\u97f3\u5167\u5bb9\u6700\u4f73\u7dad\u5ea6 4.30% 4.35% 4.40% 4.45% 4.50% 4.55% S 55\u3001T 6\u3001N 20 S 60\u3001T 6\u3001N 20 S 70\u3001T 6\u3001N 20 \u932f\u8aa4\u7387 JSEC \u8a9e\u8005\u6700\u4f73\u7dad\u5ea6 5.50% 6.00% 6.50% 7.00% 7.50% 9.00% 9.50% 10.00% MVA 35 40 u non =20x7488 =149760 (37) 45 d non =1x7488=7488 8.00% (38) 7.00% MVA 8.50% HEQ 30 u sp =20x12480=249600 (39) 6.00% 8.00% ETSI JSE 10 20 25 5.00% JSE v sp =60x12480=748800 (40) simple avg 4.00% 15 d sp =1x12480=14800 (41) complex avg 3.00% JSEC JSEC 5 g sp =8x12480=99840 2.00% (42) 6.20% 5.00% 0 1.00% \u932f\u8aa4\u7387 \u5716\u5341\u4e00\u3001complex backend JSEC \u8a9e\u8005\u6700\u4f73\u7dad\u5ea6\u6bd4\u8f03\u5716 seta setb setc avg \u53c3\u6578\u91cf\u7684\u6bd4\u4f8b 0.00% 6.19% backend \u548c complex backend \u4e8c\u7d44\u505a\u7dad\u5ea6\u7d44\u5408\u5206\u6790\u3002 \u8fa8\u8b58\u7d50\u679c\u5982\u5716\u4e03\u6240\u793a\uff1a \u5716\u4e03\u3001simple backend JSEC \u96dc\u8a0a\u74b0\u5883\u6700\u4f73\u7dad\u5ea6\u6bd4\u8f03\u5716 \u5716\u4e03\u6211\u5011\u53ef\u4ee5\u770b\u5230\u96dc\u8a0a\u7684\u6700\u4f73\u7dad\u5ea6\u662f 20 \u7dad\uff0c\u56e0\u6b64\u6211\u5011\u63a5\u8457\u56fa\u5b9a\u96dc\u8a0a 20 \u7dad\uff0c\u8a9e\u97f3\u5167\u5bb9\u4e00 \u6a23\u53d6 6 \u7dad\uff0c\u518d\u53d6\u8a9e\u8005 55 \u7dad\u300160 \u7dad\u548c 70 \u7dad\uff0c\u505a\u6e2c\u8a66\u53ef\u4ee5\u5f97\u5230\u4ee5\u4e0b\u7d50\u679c\uff1a 6.15% 6.20% 6.25% 6.30% 6.35% 6.40% JSEC \u96dc\u8a0a\u74b0\u5883\u6700\u4f73\u7dad\u5ea6 \u5716\u4e5d\u3001simple backend JSEC \u8a9e\u8005\u6700\u4f73\u7dad\u5ea6\u6bd4\u8f03\u5716 6.18% S 60\u3001T 6\u3001N 20 S 60\u3001T 8\u3001N 20 S 60\u3001T 10\u3001N 20 \u5716\u5341\u4e94\u3001simple backend MVA \u8207\u6539\u8b8a\u53c3\u6578\u91cf\u7684 JSEC \u6bd4\u4f8b\u5716 45 MVA JSE JSEC \u5716\u5341\u4e09\u3001simple backend \u5404\u7cfb\u7d71\u65b9\u6cd5\u4e0d\u540c\u74b0\u5883\u4e4b\u6bd4\u8f03\u5716 40 \u7531\u5716\u5341\u4e00\u6211\u5011\u53ef\u4ee5\u770b\u5230\u8a9e\u8005\u7684\u6700\u4f73\u7dad\u5ea6\u662f 60 \u7dad\uff0c\u56e0\u6b64\u6211\u5011\u63a5\u8457\u56fa\u5b9a\u96dc\u8a0a 20 \u7dad\uff0c\u8a9e\u8005\u53d6 60 \u7dad\uff0c\u518d\u53d6\u8a9e\u97f3\u5167\u5bb9 6 \u7dad\u30018 \u7dad\u548c 10 \u7dad\uff0c\u505a\u6e2c\u8a66\u53ef\u4ee5\u5f97\u5230\u4ee5\u4e0b\u7d50\u679c\uff1a 35 \u5716\u5341\u4e5d\u3001\u5e73\u5747\u932f\u8aa4\u7387\u6bd4\u8f03\u5716 MVA 3.4.1 Complex backend 30 30.00% 25 6.45% S 55\u3001T 6\u3001N 14 S 55\u3001T 6\u3001N 20 S 55\u3001T 6\u3001N 24 \u932f\u8aa4\u7387 \u5716\u4e5d\u6211\u5011\u53ef\u4ee5\u770b\u5230\u8a9e\u97f3\u5167\u5bb9\u7684\u6700\u4f73\u7dad\u5ea6\u662f 8 \u7dad\u3002 3.3.2 Complex backend \u7531\u65bc\u8abf\u9069\u6a21\u578b\u4e2d\u7684\u8b8a\u7570\u6578\u3001\u8f49\u79fb\u6a5f\u7387\u8207\u6b0a\u91cd\u5f71\u97ff\u5f88\u5c0f\uff0c\u56e0\u6b64\u5148\u5047\u8a2d\u8207\u6bd4\u8f03\u7684 MVA\u3001JSE \u76f8\u540c\u3002\u7dad\u5ea6\u6e2c\u8a66\u9996\u5148\u56fa\u5b9a\u8a9e\u8005(S) 55 \u7dad\uff0c\u8a9e\u97f3\u5167\u5bb9(T) 6 \u7dad\uff0c\u96dc\u8a0a(N)\u5206\u5225\u4ee5 14 \u7dad\u300120 \u7dad\u548c 24 \u7dad\u505a\u6e2c\u8a66\u5f8c\u53ef\u5f97\u4ee5\u4e0b\u7d50\u679c\uff1a seta setb setc avg MVA JSE JSEC JSEC \u96dc\u8a0a\u74b0\u5883\u6700\u4f73\u7dad\u5ea6 \u5716\u5341\u4e8c\u3001complex backend JSEC \u8a9e\u97f3\u5167\u5bb9\u6700\u4f73\u7dad\u5ea6\u6bd4\u8f03\u5716 S60 \u3001T 6\u3001N 20 S60 \u3001T 8\u3001N 20 3.80% 0.00% \u5716\u5341\u56db\u3001simple backend \u5404\u7cfb\u7d71\u65b9\u6cd5\u4e0d\u540c SNR \u4e4b\u6bd4\u8f03\u5716 4.00% 5.00% JSEC S60 \u3001T10\u3001N 20 20 15 10 5 0 avg 4.20% 10.00% 4.36% 4.37% 4.38% 4.39% 4.40% \u932f\u8aa4\u7387 25.00% MVA JSE 20 JSEC \u8a9e\u97f3\u5167\u5bb9\u6700\u4f73\u7dad\u5ea6 0.00% 5.00% 10.00% 15.00% 20.00% 45.00% 15 HEQ ETSI JSE JSEC 4.40% 10 JSEC 40.00% 5.20% 5 35.00% 5.00% MVA 0 30.00% 4.80% \u53c3\u6578\u91cf\u7684\u6bd4\u4f8b 25.00% simple backend 4.60% 15.00% JSE 20.00% complex backend</td></tr><tr><td>\u5176\u4e2d m \u70ba\u521d\u59cb\u6a21\u578b\u4e2d\u6240\u6709\u5e73\u5747\u503c\u4e32\u6210\u7684\u8d85\u5411\u91cf(super-vector)\uff1bv\u3001u\u3001d \u5206\u5225\u70ba\u7279\u5fb5\u8072\u97f3\u3001 \u7279\u5fb5\u96dc\u8a0a\u3001\u7368\u7279\u56e0\u7d20\u4e4b\u7279\u5fb5\u7a7a\u9593\uff1bvy(s)\u3001ux h (s)\u3001 dz(s)\u5206\u5225\u70ba\u4eba\u8072\u3001\u96dc\u8a0a\u3001\u7368\u7279\u56e0\u7d20\u5728 \u5404\u81ea\u7279\u5fb5\u7a7a\u9593\u7684\u5e73\u5747\u504f\u79fb\u91cf\u3002\u548c JSEC \u6700\u5927\u4e0d\u540c\u5728\u65bc\u5c11\u8003\u616e\u4e86\u8b1b\u8a71\u5167\u5bb9\u56e0\u7d20\uff0c\u6a21\u578b\u7684\u7279 \u5fb5\u5411\u91cf\u6bd4 JSEC \u9f90\u5927\u3002 3.2 \u7279\u5fb5\u7a7a\u9593\u5206\u6790 \u5716\u56db\u3001simple backend JSEC \u975e\u8a9e\u97f3\u7279\u6027\u4e4b\u96dc\u8a0a\u7279\u5fb5\u7a7a\u9593\u6295\u5f71\u5716 6.10% 6.15% 6.20% 6.25% S 55\u3001T 6\u3001N 20 S 60\u3001T 6\u3001N 20 S 70\u3001T 6\u3001N 20 \u932f\u8aa4\u7387 \u5716\u5341\u3001complex backend JSEC \u96dc\u8a0a\u74b0\u5883\u6700\u4f73\u7dad\u5ea6\u6bd4\u8f03\u5716 4.30% S 55\u3001T 6\u3001N 14 S 55\u3001T 6\u3001N 20 S 55\u3001T 6\u3001N 24 \u505a\u5be6\u9a57\u5c0d\u7167\uff0c\u4e26\u5206\u6210 simple backend \u548c complex backend \u4e8c\u7d44\u5be6\u9a57\u8a0e\u8ad6\u3002 simple backend \u6a21\u578b\u6240\u9700\u53c3\u6578 MVA JSE JSEC 10.00% JSE \u4e94\u3001\u81f4\u8b1d 3.4.1 Simple backend \u5f9e\u5716\u5341\u4e09\u548c\u5716\u5341\u56db\u6211\u5011\u767c\u73fe\uff0cJSE \u8207 JSEC \u5e73\u5747\u932f\u8aa4\u7387\u9060\u512a\u65bc MVA \u7684 7.97%\uff0c\u4f46\u662f JSEC weight 552 552 72 0.00% \u537b\u7565\u5dee JSE 0.15%\uff0c\u6211\u5011\u8a8d\u70ba JSEC \u7531\u65bc\u8072\u5b78\u6a21\u578b\u8b8a\u6210\u53ea\u6709\u4e00\u500b\u6642\uff0c\u505a simple backend mean 21528 21528 2808 variance 21528 21528 2808 5.00% \u672c\u8ad6\u6587\u6240\u9032\u884c\u5de5\u4f5c\u4e4b\u6210\u679c\uff0c\u90e8\u5206\u5728\u570b\u79d1\u6703\u5c08\u984c\u8a08\u756b\u7de8\u865f 98-2221-E-027-081-MY3 \u548c JSEC 97-2628-E-027-003-MY3 \u7684\u7d93\u8cbb\u88dc\u52a9\u4e4b\u4e0b\u9806\u5229\u5b8c\u6210\uff0c\u7279\u6b64\u81f4\u8b1d\u3002</td></tr><tr><td>\u6211\u5011\u60f3\u8981\u5f97\u77e5\u6b64\u65b9\u6cd5\u662f\u5426\u6b63\u78ba\uff0c\u80fd\u4e0d\u80fd\u6709\u6548\u5730\u5c07\u5f71\u97ff\u56e0\u7d20\u500b\u5225\u5206\u958b\uff0c\u6240\u4ee5\u5148\u4f7f\u7528 simple \u5bb9\u4e00\u6a23\u53d6 6 \u7dad\uff0c\u518d\u53d6\u8a9e\u8005 55 \u7dad\u300160 \u7dad\u548c 70 \u7dad\uff0c\u505a\u6e2c\u8a66\u53ef\u4ee5\u5f97\u5230\u4ee5\u4e0b\u7d50\u679c\uff1a \u6211\u5011\u53ef\u4ee5\u770b\u5230\u5716\u4e09\u3001\u5716\u56db\uff0c\u5728 clean \u7aef\uff0c\u96dc\u8a0a\u7279\u6027\u4e26\u4e0d\u660e\u986f\uff0c\u96a8\u8457 SNR \u589e\u52a0\uff0c\u96dc\u8a0a\u7279\u6027 \u7531\u5716\u5341\uff0c\u6211\u5011\u53ef\u4ee5\u770b\u5230\u96dc\u8a0a\u7684\u6700\u4f73\u7dad\u5ea6\u662f 20 \u7dad\uff0c\u56e0\u6b64\u6211\u5011\u63a5\u8457\u56fa\u5b9a\u96dc\u8a0a 20 \u7dad\uff0c\u8a9e\u97f3\u5167 \u7684\u5be6\u9a57\uff0c\u53ef\u80fd\u6703\u5c0e\u81f4\u6a21\u578b\u4e0d\u5920\u8907\u96dc\uff0c\u56e0\u6b64\u9020\u6210\u8fa8\u8b58\u7387\u4e0b\u964d\u3002 Transition 3598 3598 358 20 15 10 5 0 avg</td></tr></table>", | |
| "type_str": "table" | |
| } | |
| } | |
| } | |
| } |