{ "paper_id": "O15-1025", "header": { "generated_with": "S2ORC 1.0.0", "date_generated": "2023-01-19T08:10:15.221028Z" }, "title": "\u7d50\u5408 ANN \u9810\u6e2c\u3001\u5168\u57df\u8b8a\u7570\u6578\u5339\u914d\u8207\u771f\u5be6\u8ecc\u8de1\u6311\u9078\u4e4b \u57fa\u9031\u8ecc\u8de1\u7522\u751f\u65b9\u6cd5 A Pitch-contour Generation Method Combining ANN Prediction, Global Variance Matching, and Real-contour Selection", "authors": [ { "first": "Yan", "middle": [], "last": "\u53e4\u9d3b\u708e", "suffix": "", "affiliation": { "laboratory": "", "institution": "National Taiwan University of Science and Technology", "location": {} }, "email": "" }, { "first": "Kai-Wei", "middle": [], "last": "Gu", "suffix": "", "affiliation": { "laboratory": "", "institution": "National Taiwan University of Science and Technology", "location": {} }, "email": "" }, { "first": "Hao", "middle": [], "last": "Jiang", "suffix": "", "affiliation": { "laboratory": "", "institution": "National Taiwan University of Science and Technology", "location": {} }, "email": "" }, { "first": "\u570b\u7acb\u81fa\u7063\u79d1\u6280\u5927\u5b78", "middle": [], "last": "Wang", "suffix": "", "affiliation": { "laboratory": "", "institution": "National Taiwan University of Science and Technology", "location": {} }, "email": "" }, { "first": "", "middle": [], "last": "\u8cc7\u8a0a\u5de5\u7a0b\u7cfb", "suffix": "", "affiliation": { "laboratory": "", "institution": "National Taiwan University of Science and Technology", "location": {} }, "email": "" } ], "year": "", "venue": null, "identifiers": {}, "abstract": "Pitch contours are important for synthesizing highly natural speech signal. In this paper, we study a new pitch-contour generation method. The method proposed is to combine ANN prediction module with global-variance matching (GVM) and real contour selection (RCS) modules. Here, a syllable pitch contour is first analyzed and then transformed via discrete cosine transform (DCT) to a DCT-coefficient vector. Each sequence of DCT vectors analyzed from a training sentence plus contextual parameters are then used to train the ANN weights and GVM parameters. In pitch-contour generation experiments, we measure variance-ratio (VR) values for objective evaluations. The modules, GVM and RCS, are shown to be helpful to promote VR values. In addition, in subjective evaluation, the pitch-contour generation method, ANN + GVM, is shown to be more natural than the method, ANN only. Also, the method, ANN + GVM + RCS, is shown to be better than ANN + GVM.", "pdf_parse": { "paper_id": "O15-1025", "_pdf_hash": "", "abstract": [ { "text": "Pitch contours are important for synthesizing highly natural speech signal. In this paper, we study a new pitch-contour generation method. The method proposed is to combine ANN prediction module with global-variance matching (GVM) and real contour selection (RCS) modules. Here, a syllable pitch contour is first analyzed and then transformed via discrete cosine transform (DCT) to a DCT-coefficient vector. Each sequence of DCT vectors analyzed from a training sentence plus contextual parameters are then used to train the ANN weights and GVM parameters. In pitch-contour generation experiments, we measure variance-ratio (VR) values for objective evaluations. The modules, GVM and RCS, are shown to be helpful to promote VR values. In addition, in subjective evaluation, the pitch-contour generation method, ANN + GVM, is shown to be more natural than the method, ANN only. Also, the method, ANN + GVM + RCS, is shown to be better than ANN + GVM.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Abstract", "sec_num": null } ], "body_text": [ { "text": "The 2015 Conference on Computational Linguistics and Speech Processing ROCLING 2015, pp. 277-288 \uf0d3 The Association for Computational Linguistics and Chinese Language Processing \u4e00\u3001\u7dd2\u8ad6 \u4e00\u500b\u5408\u6210\u8a9e\u97f3\u4fe1\u865f\u7684\u81ea\u7136\u5ea6\u4e3b\u8981\u662f\u7531\u97fb\u5f8b\u53c3\u6578(\u5982\u57fa\u9031\u8ecc\u8de1\u3001\u97f3\u9577\u3001\u97f3\u91cf\u7b49)\u6240\u6c7a\u5b9a\uff0c\u5176 \u4e2d\u57fa\u9031\u8ecc\u8de1\u5c0d\u65bc\u81ea\u7136\u5ea6\u4e4b\u63d0\u5347\u66f4\u986f\u5f97\u91cd\u8981\uff0c\u56e0\u6b64\uff0c\u904e\u53bb\u5df2\u6709\u8a31\u591a\u4e0d\u540c\u7684\u97f3\u7bc0\u57fa\u9031\u8ecc\u8de1\u7522 \u751f\u65b9\u6cd5\u88ab\u5148\u524d\u7684\u7814\u7a76\u8005\u6240\u63d0\u51fa [1, 2, 3, 4, 5, 6] \u3002\u76ee\u524d\uff0c\u96b1\u85cf\u5f0f\u99ac\u53ef\u592b\u6a21\u578b(hidden Markov model, HMM) \u96d6\u7136\u5df2\u88ab\u8a31\u591a\u4eba\u63a1\u7528\u65bc\u4f5c\u8a9e\u97f3\u5408\u6210\u7684\u7814\u7a76 [7, 8] \uff0c\u7136\u800c MSD-HMM (multi-space probability distribution HMM)\u7522\u751f\u51fa\u7684\u57fa\u9031\u8ecc\u8de1\u4e26\u4e0d\u5341\u5206\u5730\u4ee4\u4eba\u6eff\u610f\uff0c\u9019\u7a2e \u60c5\u5f62\u5df2\u6709\u4e0d\u5c11\u4eba\u6ce8\u610f\u5230 [3, 6] ", "cite_spans": [ { "start": 275, "end": 278, "text": "[1,", "ref_id": "BIBREF0" }, { "start": 279, "end": 281, "text": "2,", "ref_id": "BIBREF1" }, { "start": 282, "end": 284, "text": "3,", "ref_id": "BIBREF2" }, { "start": 285, "end": 287, "text": "4,", "ref_id": "BIBREF3" }, { "start": 288, "end": 290, "text": "5,", "ref_id": "BIBREF4" }, { "start": 291, "end": 293, "text": "6]", "ref_id": "BIBREF5" }, { "start": 352, "end": 355, "text": "[7,", "ref_id": "BIBREF6" }, { "start": 356, "end": 358, "text": "8]", "ref_id": "BIBREF7" }, { "start": 445, "end": 448, "text": "[3,", "ref_id": "BIBREF2" }, { "start": 449, "end": 451, "text": "6]", "ref_id": "BIBREF5" } ], "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": "( ) (0) ( 1) ( 1) 2 ( ) cos( ), 0,1,..., 23 N m m k k N c m x x N x k m \u03c0 \u2212 \u22c5 \u22c5 = \u2212 = + \u2212 \u22c5 \u2212 + \u22c5 \u22c5 = \uf0e5 (1) \u5176\u4e2d x(k)\u8868\u793a\u4e00\u500b\u97f3\u7bc0\u57fa\u9031\u8ecc\u8de1\u7684\u7b2c k \u500b\u97f3\u6846\u7684\u57fa\u983b\u503c(\u4ee5 Hz \u70ba\u55ae\u4f4d)\uff0cc(m)\u8868\u793a DCT \u8f49\u63db\u5f8c\u7684\u7b2c m \u968e\u4fc2\u6578\uff0c\u800c N \u5247\u662f\u8a72\u97f3\u7bc0\u7684\u97f3\u6846\u6578\u3002 \u5c0d\u65bc\u516c\u5f0f(1)\uff0c\u5176\u5c0d\u61c9\u7684 DCT \u53cd\u8f49\u63db\u516c\u5f0f\u70ba\uff1a 2 1 1 2( 1) 1 ( ) (0) ( 1) ( 1) 2 ( ) cos( ) , 0,1, ..., 1 M k k m m N M x k c c M c m k N \u03c0 \u2212 \u22c5 \u22c5 = \u2212 \u2212 \uf0e9 \uf0f9 = + \u2212 \u22c5 \u2212 + \u22c5 \u22c5 \uf0ea \uf0fa \uf0eb \uf0fb = \u2212 \uf0e5", "eq_num": "(2)" } ], "section": "", "sec_num": null }, { "text": "EQUATION", "cite_spans": [], "ref_spans": [], "eq_spans": [ { "start": 0, "end": 8, "text": "EQUATION", "ref_id": "EQREF", "raw_str": "( ) 2 1 ( ( ) ) ( ) , n k k k k i i i j v c j m n k = \uf0e9 \uf0f9 = \u2212 \uf0ea \uf0fa \uf0eb \uf0fb \uf0e5 (3) \u5176\u4e2d n(k)\u8868\u793a\u7b2c k \u8a9e\u53e5\u88e1\u7684\u97f3\u7bc0\u500b\u6578\uff0c ( ) k i c j \u8868\u793a\u7b2c j \u500b\u97f3\u7bc0\u57fa\u9031\u8ecc\u8de1 DCT \u5411\u91cf\u4e4b\u7b2c i \u7dad\u7684\u4fc2\u6578\uff0c\u800c k i m \u8868\u793a ( ) k i c j , j=1, \u2026, n(k)\u7684\u5e73\u5747\u503c\u3002 \u5982\u6b64\uff0c\u6a6b\u8de8 750 \u53e5\u8a13\u7df4\u8a9e\u53e5\u7684 DCT \u5411\u91cf\u7b2c i \u7dad\u4e4b\u5168\u57df\u8b8a\u7570\u6578\uff0c\u5c31\u53ef\u4ee5\u516c\u5f0f(4)\u4f86\u4f5c\u8a08\u7b97\uff1a 1 1 , N k i i k N g v = = \uf0e5", "eq_num": "(4" } ], "section": "", "sec_num": null }, { "text": "EQUATION", "cite_spans": [], "ref_spans": [], "eq_spans": [ { "start": 0, "end": 8, "text": "EQUATION", "ref_id": "EQREF", "raw_str": "i i i i i i c c m w g v m i \uf0e9 \uf0f9 = \u2212 \u22c5 + + = \uf0eb \uf0fb", "eq_num": "(5)" } ], "section": "", "sec_num": null }, { "text": ", d k d k L D L D k d VR \u03c3 \u03c3 = = = \u22c5 \uf0e5 \uf0e5 (6) \u5176\u4e2d L \u8868\u793a\u570b\u8a9e\u97fb\u6bcd\u7684\u985e\u5225\u6578(\u5728\u6b64 L=36)\uff1b D \u8868\u793a\u57fa\u9031\u8ecc\u8de1 DCT \u5411\u91cf\u7684\u7dad\u5ea6\u6578\uff1b\u02c6d k \u03c3 \u8868 \u793a\u7a0b\u5f0f\u7522\u751f\u51fa\u7684\u57fa\u9031\u8ecc\u8de1 DCT \u5411\u91cf\u4e4b\u4e2d\uff0c\u628a\u5c6c\u65bc\u7b2c k \u985e\u97fb\u6bcd\u4e4b DCT \u5411\u91cf\u7b2c d \u7dad\u7684\u4fc2\u6578 \u62ff\u53bb\u8a08\u7b97\u51fa\u7684\u8b8a\u7570\u6578\uff1b d k \u03c3 \u5247\u8868\u793a\u539f\u59cb\u97f3\u7bc0\u8a9e\u6599\u5206\u6790\u51fa\u7684\u57fa\u9031\u8ecc\u8de1 DCT \u5411\u91cf\u4e4b\u4e2d\uff0c\u628a\u5c6c \u65bc\u7b2c k \u985e\u97fb\u6bcd\u4e4b DCT \u5411\u91cf\u7b2c d \u7dad\u7684\u4fc2\u6578\u62ff\u53bb\u8a08\u7b97\u51fa\u7684\u8b8a\u7570\u6578\u3002\u9700\u6ce8\u610f\u7684\u662f\uff0c\u5728\u6b64 D \u7684 \u503c\u70ba 23\uff0c\u56e0\u70ba ANN \u7522\u751f\u7684 DCT \u4fc2\u6578 c0\uff0c\u6211\u5011\u4e26\u672a\u5c0d\u5b83\u4f5c GVM \u8abf\u6574\uff0c\u4e5f\u672a\u628a\u5b83\u53d6\u4ee3 \u6210 RCS \u9078\u51fa\u7684 DCT \u5411\u91cf\u4e4b", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null } ], "back_matter": [], "bib_entries": { "BIBREF0": { "ref_id": "b0", "title": "An RNN-based prosodic information synthesizer for Mandarin text-to-speech", "authors": [ { "first": "S", "middle": [ "H" ], "last": "Chen", "suffix": "" }, { "first": "S", "middle": [ "H" ], "last": "Hwang", "suffix": "" }, { "first": "Y", "middle": [ "R" ], "last": "Wang", "suffix": "" } ], "year": 1998, "venue": "IEEE trans. Speech and Audio Processing", "volume": "6", "issue": "3", "pages": "226--239", "other_ids": {}, "num": null, "urls": [], "raw_text": "S. H. Chen, S. H. Hwang, and Y. R. Wang, \"An RNN-based prosodic information synthesizer for Mandarin text-to-speech\", IEEE trans. Speech and Audio Processing, Vol. 6, No. 3, pp. 226-239, 1998.", "links": null }, "BIBREF1": { "ref_id": "b1", "title": "A novel prosodic-information synthesizer based on recurrent fuzzy neural network for the Chinese TTS system", "authors": [ { "first": "C", "middle": [ "T" ], "last": "Lin", "suffix": "" }, { "first": "R", "middle": [ "C" ], "last": "Wu", "suffix": "" }, { "first": "J", "middle": [ "Y" ], "last": "Chang", "suffix": "" }, { "first": "S", "middle": [ "F" ], "last": "Liang", "suffix": "" } ], "year": 2004, "venue": "IEEE trans. Systems, Man, and Cybernetics", "volume": "34", "issue": "1", "pages": "309--324", "other_ids": {}, "num": null, "urls": [], "raw_text": "C. T. Lin, R. C. Wu, J. Y. Chang, and S. F. Liang, \"A novel prosodic-information synthesizer based on recurrent fuzzy neural network for the Chinese TTS system\", IEEE trans. Systems, Man, and Cybernetics, Vol. 34, No. 1, pp. 309-324, 2004.", "links": null }, "BIBREF2": { "ref_id": "b2", "title": "Exploiting prosody hierarchy and dynamic features for pitch modeling and generation in HMM-based speech synthesis", "authors": [ { "first": "C", "middle": [ "C" ], "last": "Hsia", "suffix": "" }, { "first": "C", "middle": [ "H" ], "last": "Wu", "suffix": "" }, { "first": "J", "middle": [ "Y" ], "last": "Wu", "suffix": "" } ], "year": 1994, "venue": "IEEE trans. Audio, Speech, and Language Processing", "volume": "18", "issue": "8", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "C. C. Hsia, C. H. Wu, and J. Y. Wu, \"Exploiting prosody hierarchy and dynamic features for pitch modeling and generation in HMM-based speech synthesis\", IEEE trans. Audio, Speech, and Language Processing, Vol. 18, No. 8, pp. 1994-2003, 2010.", "links": null }, "BIBREF3": { "ref_id": "b3", "title": "An HMM based pitch-contour generation method for Mandarin speech synthesis", "authors": [ { "first": "H", "middle": [ "Y" ], "last": "Gu", "suffix": "" }, { "first": "C", "middle": [ "C" ], "last": "Yang", "suffix": "" } ], "year": 2011, "venue": "Journal of Information Science and Engineering", "volume": "27", "issue": "5", "pages": "1561--1580", "other_ids": {}, "num": null, "urls": [], "raw_text": "H. Y. Gu and C. C. Yang, \"An HMM based pitch-contour generation method for Mandarin speech synthesis\", Journal of Information Science and Engineering, Vol. 27, No. 5, pp. 1561-1580, 2011.", "links": null }, "BIBREF4": { "ref_id": "b4", "title": "Pitch contour model for Chinese text-to-speech using CART and statistical model", "authors": [ { "first": "M", "middle": [], "last": "Dong", "suffix": "" }, { "first": "K", "middle": [ "T" ], "last": "Lua", "suffix": "" } ], "year": 2002, "venue": "Int. Conf. on Spoken Language Processing", "volume": "", "issue": "", "pages": "2405--2408", "other_ids": {}, "num": null, "urls": [], "raw_text": "M. Dong, K. T. Lua, \"Pitch contour model for Chinese text-to-speech using CART and statistical model,\" Int. Conf. on Spoken Language Processing, Denver, USA, pp. 2405-2408, 2002.", "links": null }, "BIBREF5": { "ref_id": "b5", "title": "Improving F0 prediction using bidirectional associative memories and syllable-level F0 features for HMM-based Mandarin speech synthesis", "authors": [ { "first": "L", "middle": [], "last": "Gao", "suffix": "" }, { "first": "Z", "middle": [ "H" ], "last": "Ling", "suffix": "" }, { "first": "L", "middle": [ "H" ], "last": "Chen", "suffix": "" }, { "first": "L", "middle": [ "R" ], "last": "Dai", "suffix": "" } ], "year": 2014, "venue": "Proceeding of ISCSLP", "volume": "", "issue": "", "pages": "275--279", "other_ids": {}, "num": null, "urls": [], "raw_text": "L. Gao, Z. H. Ling, L. H. Chen, and L. R. Dai, \"Improving F0 prediction using bidirectional associative memories and syllable-level F0 features for HMM-based Mandarin speech synthesis\", Proceeding of ISCSLP, Singapore, pp. 275-279, 2014.", "links": null }, "BIBREF6": { "ref_id": "b6", "title": "Simultaneous modeling of spectrum, pitch and duration in HMM based speech synthesis", "authors": [ { "first": "T", "middle": [], "last": "Yoshimura", "suffix": "" }, { "first": "K", "middle": [], "last": "Tokuda", "suffix": "" }, { "first": "T", "middle": [], "last": "Masuko", "suffix": "" }, { "first": "T", "middle": [], "last": "Kobayashi", "suffix": "" }, { "first": "T", "middle": [], "last": "Kitamura", "suffix": "" } ], "year": 1999, "venue": "Proceeding of EUROSPEECH", "volume": "", "issue": "", "pages": "2347--2350", "other_ids": {}, "num": null, "urls": [], "raw_text": "T. Yoshimura, K. Tokuda, T. Masuko, T. Kobayashi, and T. Kitamura, \"Simultaneous modeling of spectrum, pitch and duration in HMM based speech synthesis\", Proceeding of EUROSPEECH, Budapest, Hungary, pp. 2347-2350, 1999.", "links": null }, "BIBREF7": { "ref_id": "b7", "title": "The HMM-based speech synthesis system (HTS) version 2.0", "authors": [ { "first": "H", "middle": [], "last": "Zen", "suffix": "" }, { "first": "T", "middle": [], "last": "Nose", "suffix": "" }, { "first": "J", "middle": [], "last": "Yamagishi", "suffix": "" }, { "first": "S", "middle": [], "last": "Sako", "suffix": "" }, { "first": "T", "middle": [], "last": "Masuko", "suffix": "" }, { "first": "A", "middle": [ "W" ], "last": "Black", "suffix": "" }, { "first": "K", "middle": [], "last": "Tokuda", "suffix": "" } ], "year": 2007, "venue": "Proceeding of 6-th ISCA Workshop on Speech Synthesis", "volume": "", "issue": "", "pages": "294--299", "other_ids": {}, "num": null, "urls": [], "raw_text": "H. Zen, T. Nose, J. Yamagishi, S. Sako, T. Masuko, A. W. Black, and K. Tokuda, \"The HMM-based speech synthesis system (HTS) version 2.0\", Proceeding of 6-th ISCA Workshop on Speech Synthesis, Bonn, Germany, pp. 294-299, 2007.", "links": null }, "BIBREF8": { "ref_id": "b8", "title": "A speech parameter generation algorithm considering global variance for HMM-based speech synthesis", "authors": [ { "first": "T", "middle": [], "last": "Toda", "suffix": "" }, { "first": "K", "middle": [], "last": "Tokuda", "suffix": "" } ], "year": 2007, "venue": "IEICE trans. INF. & SYST", "volume": "", "issue": "5", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "T. Toda and K. Tokuda, \"A speech parameter generation algorithm considering global variance for HMM-based speech synthesis\", IEICE trans. INF. & SYST., Vol. E90-D, No. 5, May 2007.", "links": null }, "BIBREF9": { "ref_id": "b9", "title": "Towards a voice conversion system based on frame selection", "authors": [ { "first": "T", "middle": [], "last": "Dutoit", "suffix": "" }, { "first": "A", "middle": [], "last": "Holzapfel", "suffix": "" }, { "first": "M", "middle": [], "last": "Jottrand", "suffix": "" }, { "first": "A", "middle": [], "last": "Moinet", "suffix": "" }, { "first": "J", "middle": [], "last": "Perez", "suffix": "" }, { "first": "Y", "middle": [], "last": "Stylianou", "suffix": "" } ], "year": 2007, "venue": "Int. Conf. Acoustics, Speech, and signal Processing", "volume": "", "issue": "", "pages": "513--516", "other_ids": {}, "num": null, "urls": [], "raw_text": "T. Dutoit, A. Holzapfel, M. Jottrand, A. Moinet, J. Perez, and Y. Stylianou, \"Towards a voice conversion system based on frame selection\", Int. Conf. Acoustics, Speech, and signal Processing, Honolulu, Hawaii, pp. 513-516, 2007.", "links": null }, "BIBREF10": { "ref_id": "b10", "title": "Descrete-time Signal Processing", "authors": [ { "first": "A", "middle": [ "V" ], "last": "Oppenheim", "suffix": "" }, { "first": "R", "middle": [ "W" ], "last": "Schafter", "suffix": "" } ], "year": 1999, "venue": "", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "A. V. Oppenheim and R. W. Schafter, Descrete-time Signal Processing, second ed., Prentice-Hall, 1999.", "links": null }, "BIBREF11": { "ref_id": "b11", "title": "A system framework for integrated synthesis of Mandarin, Min-nan, and Hakka speech", "authors": [ { "first": "H", "middle": [ "Y" ], "last": "Gu", "suffix": "" }, { "first": "Y", "middle": [ "Z" ], "last": "Zhou", "suffix": "" }, { "first": "H", "middle": [ "L" ], "last": "Liau", "suffix": "" } ], "year": 2007, "venue": "Int. Journal of Computational Linguistics and Chinese Language Processing", "volume": "12", "issue": "4", "pages": "371--390", "other_ids": {}, "num": null, "urls": [], "raw_text": "H. Y. Gu, Y. Z. Zhou, and H. L. Liau, \"A system framework for integrated synthesis of Mandarin, Min-nan, and Hakka speech\", Int. Journal of Computational Linguistics and Chinese Language Processing, Vol. 12, No. 4, pp. 371-390, 2007.", "links": null }, "BIBREF12": { "ref_id": "b12", "title": "Voice conversion using dynamic frequency warping with amplitude scaling, for parallel or nonparallel corpora", "authors": [ { "first": "E", "middle": [], "last": "Godoy", "suffix": "" }, { "first": "O", "middle": [], "last": "Rosec", "suffix": "" }, { "first": "T", "middle": [], "last": "Chonavel", "suffix": "" } ], "year": 2012, "venue": "IEEE trans. Audio, Speech, and Language Processing", "volume": "20", "issue": "", "pages": "1313--1323", "other_ids": {}, "num": null, "urls": [], "raw_text": "E. Godoy, O. Rosec, and T. Chonavel, \"Voice conversion using dynamic frequency warping with amplitude scaling, for parallel or nonparallel corpora\", IEEE trans. Audio, Speech, and Language Processing, Vol. 20, No. 4, pp. 1313-1323, 2012.", "links": null }, "BIBREF13": { "ref_id": "b13", "title": "Speech synthesis using articulatory-knowledge based HMM structure", "authors": [ { "first": "H", "middle": [ "Y" ], "last": "Gu", "suffix": "" }, { "first": "M", "middle": [ "Y" ], "last": "Lai", "suffix": "" }, { "first": "W", "middle": [ "S" ], "last": "Hong", "suffix": "" } ], "year": 2014, "venue": "Int. Conf. on Machine Learning and Cybernetics", "volume": "", "issue": "", "pages": "371--376", "other_ids": {}, "num": null, "urls": [], "raw_text": "H. Y. Gu, M. Y. Lai, and W. S. Hong, \"Speech synthesis using articulatory-knowledge based HMM structure\", Int. Conf. on Machine Learning and Cybernetics, Lanzhou, China, pp. 371-376, 2014.", "links": null } }, "ref_entries": { "TABREF0": { "html": null, "num": null, "content": "
\u4e00\u6b65\u63d0\u5347\u6240\u7522\u751f\u7684\u57fa\u9031\u8ecc\u8de1\u7684\u81ea\u7136\u5ea6\u3002\u95dc\u65bc RCS \u7684\u5be6\u4f5c\uff0c\u6211\u5011\u53ef\u4f9d\u64da\u5404\u500b\u97f3\u7bc0\u7684\u8a9e\u5883 \u8cc7\u6599\u4f86\u4f5c\u8a9e\u5883\u7684\u5206\u985e\uff0c\u7136\u5f8c\u628a\u5c6c\u65bc\u4e0d\u540c\u8a9e\u5883\u5206\u985e\u7684\u5404\u500b\u771f\u5be6\u57fa\u9031\u8ecc\u8de1 DCT \u5411\u91cf\uff0c\u5206\u5225 \u653e\u5165\u4e0d\u540c\u7684\u6536\u96c6\u5340(pool)\u88e1\u3002 \u6574\u9ad4\u4f86\u8aaa\uff0c\u6211\u5011\u7cfb\u7d71\u5728\u8a13\u7df4\u968e\u6bb5\u7684\u8655\u7406\u6d41\u7a0b\u5982\u5716\u4e00\u6240\u793a\u3002\u9996\u5148\u5c0d\u6bcf\u500b\u9304\u97f3\u8a9e\u53e5\u7684\u7684\u5404\u500b \u97f3\u7bc0\u4f5c\u57fa\u9031\u8ecc\u8de1\u5206\u6790\uff1b\u63a5\u8457\uff0c\u628a\u5404\u500b\u97f3\u7bc0\u7684\u57fa\u9031\u8ecc\u8de1\u8f49\u63db\u6210\u56fa\u5b9a\u7dad\u5ea6\u7684 DCT \u4fc2\u6578\u5411\u91cf\uff1b \u7136\u5f8c\u62ff\u5404\u500b\u8a13\u7df4\u8a9e\u53e5\u7684 DCT \u5411\u91cf\u5e8f\u5217\u53ca\u5404\u97f3\u7bc0\u5c0d\u61c9\u7684\u8a9e\u5883\u8cc7\u6599\uff0c\u53bb\u8a13\u7df4 ANN \u70ba\u57fa\u790e \u7684\u57fa\u9031\u8ecc\u8de1\u7522\u751f\u6a21\u578b\u3002\u9664\u4e86\u8a13\u7df4 ANN \u6a21\u578b\u4e4b\u5916\uff0c\u6211\u5011\u4e5f\u5c0d\u5404\u500b\u8a13\u7df4\u8a9e\u53e5\u7684 DCT \u5411\u91cf \u5e8f\u5217\u4f5c\u5206\u6790\uff0c\u4ee5\u6c42\u5f97 GVM \u5339\u914d\u6240\u9700\u7684\u53c3\u6578\u3002\u6b64\u5916\uff0c\u6211\u5011\u4f9d\u64da\u5404\u500b\u97f3\u7bc0\u7684\u8a9e\u5883\u5206\u985e\uff0c\u628a \u5b83\u7684\u57fa\u9031\u8ecc\u8de1 DCT \u5411\u91cf\u653e\u5165\u5c0d\u61c9\u7684\u6536\u96c6\u5340\u88e1\u3002 \u53e6\u4e00\u65b9\u9762\uff0c\u7522\u751f\u57fa\u9031\u8ecc\u8de1\u7684\u6574\u9ad4\u6d41\u7a0b\u5982\u5716\u4e8c\u6240\u793a\u3002\u9996\u5148\u8f38\u5165\u4e00\u500b\u6587\u53e5\uff0c\u63a5\u8457\u7d93\u7531\u641c\u5c0b\u8a5e \u5178\u4f86\u78ba\u8a8d\u5404\u500b\u4e2d\u6587\u5b57\u7684\u97f3\u7bc0\u767c\u97f3\u8207\u97f3\u8abf\uff1b\u4f9d\u64da\u67e5\u8a62\u51fa\u7684\u4e00\u5e8f\u5217\u97f3\u7bc0\u767c\u97f3\u8207\u97f3\u8abf\uff0c\u5c31\u53ef\u70ba \u5404\u97f3\u7bc0\u6e96\u5099\u5b83\u5c0d\u61c9\u7684\u8a9e\u5883\u53c3\u6578\uff0c\u7136\u5f8c\u5c07\u5404\u97f3\u7bc0\u7684\u8a9e\u5883\u53c3\u6578\u8f38\u5165 ANN \u6a21\u578b\uff0c\u53bb\u9810\u6e2c\u8a72\u97f3 Detect pitch contour Training sentences Transform to DCT coeff. (each syllable's contour) Train ANN Context parameters ANN weights GV analysis GVM parameters Collect real contours real-contour pools (each context type) end \u5716\u4e00\u3001\u57fa\u9031\u8ecc\u8de1\u6a21\u578b\u4e4b\u53c3\u6578\u8a13\u7df4\u7684\u4e3b\u6d41\u7a0b start Input a written sentence Dictionary Determine contextual parameters (for each syllable) ANN prediction ANN weights param. Real contour pools Real contour selection Genertd. pitch contour (each syllable) end \u5716\u4e8c\u3001\u57fa\u9031\u8ecc\u8de1\u7522\u751f\u968e\u6bb5\u4e4b\u4e3b\u6d41\u7a0b \u4e8c\u3001\u6a21\u578b\u53c3\u6578\u8a13\u7df4 \u5982\u5716\u4e00\u6240\u793a\uff0c\u6211\u5011\u9700\u8981\u8a13\u7df4 ANN \u6a21\u578b\u7684\u6b0a\u91cd\uff0c\u5206\u6790\u51fa GVM \u5339\u914d\u6240\u9700\u7684\u53c3\u6578\uff0c\u53ca\u5206\u5225 \u5132\u5b58\u4e0d\u540c\u8a9e\u5883\u985e\u578b\u7684\u771f\u5be6\u57fa\u9031\u8ecc\u8de1(\u5373 DCT \u4fc2\u6578\u5411\u91cf)\u3002 (\u4e00) \u3001\u8a9e\u53e5\u9304\u97f3\u8207\u57fa\u9031\u8ecc\u8de1\u5075\u6e2c \u5728\u6b64\u7814\u7a76\u4e2d\uff0c\u6211\u5011\u9080\u8acb\u4e86\u4e00\u4f4d\u7537\u6027\u8a9e\u8005\u65bc\u9304\u97f3\u5ba4\u4e2d\u9304\u88fd\u4e86 810 \u53e5\u8a9e\u53e5\uff0c\u800c\u7e3d\u97f3\u7bc0\u6578\u70ba 7,161 \u500b\u97f3\u7bc0\u3002\u5728\u9304\u97f3\u4e4b\u5f8c\uff0c\u5148\u4ee5 HTK (HMM toolkit) \u8edf\u9ad4\u9032\u884c\u81ea\u52d5\u6a19\u97f3\uff0c\u518d\u4f7f\u7528 WaveSurfer \u8edf\u9ad4\u4f86\u5c0d\u5404\u97f3\u7bc0\u7684\u6642\u9593\u908a\u754c\u4f5c\u4eba\u5de5\u5fae\u8abf\u3002 \u95dc\u65bc\u97f3\u7bc0\u57fa\u9031\u8ecc\u8de1\u4e4b\u5075\u6e2c\uff0c\u6211\u5011\u4f7f\u7528 HTS (HMM-based speech synthesis system)\u8edf\u9ad4\u5167\u542b \u7684 SPTK (Speech Signal Processing Toolkit)\u6a21\u7d44[8]\u4f86\u9032\u884c\uff0c\u4e26\u4e14\u8a2d\u5b9a\u97f3\u6a94\u7684\u53d6\u6a23\u7387\u8a2d\u70ba 22,050 Hz\uff0c\u800c\u97f3\u6846\u4f4d\u79fb\u5247\u8a2d\u70ba 110 \u500b\u6a23\u672c\u9ede\u3002\u5728\u81ea\u52d5\u5075\u6e2c\u57fa\u9031\u8ecc\u8de1\u4e4b\u5f8c\uff0c\u6211\u5011\u767c\u73fe\u6709 \u8a31\u591a\u97f3\u6846\u6240\u5075\u6e2c\u51fa\u7684\u57fa\u983b\u503c\u662f\u932f\u8aa4\u7684\uff0c\u4f8b\u5982\u4e00\u500b\u6709\u8072(voiced)\u97f3\u6846\u7684\u57fa\u983b\u503c\u53ef\u80fd\u88ab\u5075\u6e2c \u70ba 0\uff0c\u5373\u8aa4\u5224\u70ba\u7121\u8072(unvoiced)\uff0c\u6216\u662f\u88ab\u5075\u6e2c\u6210\u771f\u5be6\u983b\u7387\u7684\u4e00\u534a\u6216\u5169\u500d\u7684\u60c5\u5f62\u3002\u56e0\u6b64\u6211\u5011 \u64b0\u5beb\u4e86\u4e00\u500b\u5de5\u5177\u7a0b\u5f0f\uff0c\u4f86\u5c0d\u5075\u6e2c\u932f\u8aa4\u7684\u57fa\u9031\u8ecc\u8de1\u4f5c\u534a\u81ea\u52d5\u6216\u662f\u624b\u52d5\u7684\u66f4\u6b63\u8655\u7406\u3002 (\u4e8c) \u3001\u96e2\u6563\u9918\u5f26\u8f49\u63db \u7531\u65bc\u4e00\u500b\u8a9e\u53e5\u4e2d\u5404\u97f3\u7bc0\u7684\u57fa\u9031\u8ecc\u8de1\u9577\u5ea6\u4e0d\u4e00\uff0c\u9577\u5ea6\u53ef\u80fd\u4ecb\u65bc 30 \u81f3 80 \u500b\u97f3\u6846\u4e4b\u9593\uff0c\u70ba\u4e86 \u628a\u57fa\u9031\u8ecc\u8de1\u8868\u793a\u6210\u56fa\u5b9a\u7dad\u5ea6\u6578\u7684\u8cc7\u6599\uff0c\u6211\u5011\u9078\u64c7\u4ee5\u96e2\u6563\u9918\u5f26\u8f49\u63db(DCT)\u4e4b\u4fc2\u6578\u4f86\u8868\u793a\u5404 \u97f3\u7bc0\u7684\u57fa\u9031\u8ecc\u8de1\u3002\u81f3\u65bc\u7dad\u5ea6\u6578\u91cf\u4e4b\u9078\u64c7\uff0c\u5728\u6bd4\u8f03\u904e\u591a\u7a2e\u7dad\u5ea6\u6578\u4e4b DCT \u8f49\u63db\u8207\u53cd\u8f49\u63db\u56de \u4f86\u7684\u57fa\u9031\u8ecc\u8de1\u66f2\u7dda\u5f8c\uff0c\u6211\u5011\u6c7a\u5b9a\u5c07\u7dad\u5ea6\u6578\u8a2d\u70ba 24\u3002\u4e00\u500b\u539f\u59cb\u7684\u57fa\u9031\u8ecc\u8de1\u3001\u548c DCT \u53cd\u8f49 \u63db\u56de\u4f86\u4e4b\u66f2\u7dda\u4f8b\u5b50\u5982\u5716\u4e09\u6240\u793a\uff0c\u6211\u5011\u89ba\u5f97 16 \u968e DCT \u53cd\u8f49\u63db\u6240\u5f97\u4e4b\u66f2\u7dda\uff0c\u4ecd\u4e0d\u5920\u5fe0\u5be6\u65bc \u539f\u59cb\u66f2\u7dda\u3002 \u8a73\u7d30\u4f86\u8aaa\uff0c\u672c\u7814\u7a76\u88e1\u4f7f\u7528\u7684\u662f DCT-I \u4e4b\u96e2\u6563\u9918\u5f26\u8f49\u63db[11]\uff0c\u5176\u6b63\u5411\u8f49\u63db\u4e4b\u516c\u5f0f\u70ba\uff1a 2 \u7bc0\u7684\u57fa\u9031\u8ecc\u8de1(\u5373 start GVM GV matching 1 1
", "text": "\u6211\u5011\u89ba\u5f97\u57fa\u9031\u8ecc\u8de1\u4e4b\u7522\u751f\uff0c\u4e26\u4e0d\u9700\u8981\u548c\u983b\u8b5c\u4fc2\u6578\u4e4b\u7522\u751f\u7d81\u5728\u540c\u4e00\u7a2e\u6a5f\u5236(\u5373 HMM)\u88e1\uff0c \u4e26\u4e14\u6211\u5011\u60f3\u8981\u9032\u4e00\u6b65\u63d0\u5347\u6240\u7522\u751f\u51fa\u7684\u57fa\u9031\u8ecc\u8de1\u7684\u81ea\u7136\u5ea6\uff0c\u56e0\u6b64\u5728\u672c\u8ad6\u6587\u4e2d\uff0c\u6211\u5011\u5617\u8a66\u7814 \u7a76\u3001\u63d0\u51fa\u4e00\u7a2e\u628a\u985e\u795e\u7d93\u7db2\u8def(artificial neural network, ANN)\u9810\u6e2c[1, 2]\u3001\u5168\u57df\u8b8a\u7570\u6578\u5339\u914d (global-variance matching, GVM)\u8207\u771f\u5be6\u8ecc\u8de1\u6311\u9078(real contour selection, RCS)\u4e09\u8005\u4f5c\u7d50\u5408 \u7684\u65b9\u6cd5\uff0c\u5e0c\u671b\u7528\u4ee5\u63d0\u5347\u5408\u6210\u8a9e\u97f3\u7684\u81ea\u7136\u5ea6\u3002 \u904e\u53bb Toda \u8207 Tokuda \u63d0\u51fa GVM \u4e4b\u4f5c\u6cd5[9]\uff0c\u4f86\u5c0d HMM \u7522\u751f\u7684\u983b\u8b5c\u4fc2\u6578\u4f5c\u8abf\u6574\uff0c\u4ee5\u6e1b\u7de9 \u767c\u751f\u983b\u8b5c\u904e\u5ea6\u5e73\u6ed1(spectral over-smoothing)\u7684\u73fe\u8c61\uff0c\u800c\u85c9\u4ee5\u63d0\u5347\u5408\u6210\u8a9e\u7684\u97f3\u8cea\u3002\u5728\u6b64\uff0c \u6211\u5011\u767c\u73fe\u5230 ANN \u7522\u751f\u7684\u8868\u793a\u57fa\u9031\u8ecc\u8de1\u7684 DCT (discrete cosine transform)\u4fc2\u6578\uff0c\u4e5f\u540c\u6a23\u6703 \u767c\u751f\u904e\u5ea6\u5e73\u6ed1(over smoothing)\u7684\u73fe\u8c61\uff0c\u56e0\u6b64\u6211\u5011\u89ba\u5f97\u5c0d ANN \u7522\u751f\u7684\u57fa\u9031\u8ecc\u8de1 DCT \u4fc2 \u6578\uff0c\u4f5c GVM \u5339\u914d\u5c07\u6709\u52a9\u65bc\u63d0\u5347 ANN \u57fa\u9031\u8ecc\u8de1\u7684\u81ea\u7136\u5ea6\u3002\u6b64\u5916\uff0c\u6211\u5011\u4e5f\u53d7\u5230\u53e6\u4e00\u500b\u89c0 \u5ff5\u7684\u555f\u767c\uff0c\u5c31\u662f\u8a9e\u97f3\u8f49\u63db(voice conversion)\u9818\u57df\u4e2d\u524d\u4eba\u63d0\u51fa\u7684\uff0c\u4ee5\u6311\u9078\u76ee\u6a19\u8a9e\u8005\u97f3\u6846\u7684 \u771f\u5be6\u983b\u8b5c\u4fc2\u6578\u4f86\u53d6\u4ee3\u8f49\u63db\u51fa\u7684\u983b\u8b5c\u4fc2\u6578\uff0c\u5982\u6b64\u7528\u4ee5\u6539\u9032\u8f49\u63db\u51fa\u8a9e\u97f3\u7684\u97f3\u8cea[10]\u3002\u56e0\u6b64\uff0c \u6211\u5011\u8a8d\u70ba\u628a ANN \u7522\u751f\u4e26\u4e14\u7d93\u904e GVM \u5339\u914d\u7684 DCT \u4fc2\u6578\u5411\u91cf X \u4f5c\u70ba\u53c3\u8003\uff0c\u800c\u64da\u4ee5\u9078\u51fa \u4e00\u500b\u6700\u9760\u8fd1 X \u7684\u771f\u5be6\u57fa\u9031\u8ecc\u8de1 DCT \u4fc2\u6578\u5411\u91cf Y\uff0c\u7136\u5f8c\u628a Y \u62ff\u53bb\u53d6\u4ee3 X\uff0c\u5982\u6b64\u5c07\u53ef\u66f4\u9032 DCT \u4fc2\u6578)\uff1b\u5c0d\u65bc ANN \u9810\u6e2c\u51fa\u7684\u57fa\u9031\u8ecc\u8de1\uff0c\u63a5\u8457\u4f7f\u7528\u8a13\u7df4\u968e\u6bb5\u5132\u5b58\u7684 \u5168\u57df\u8b8a\u7570\u6578(GV)\u53c3\u6578\u53bb\u5c0d DCT \u4fc2\u6578\u9032\u884c GVM \u5339\u914d\uff1b\u4e4b\u5f8c\uff0c\u4f9d\u64da GVM \u5339\u914d\u8abf\u6574\u904e\u7684 \u57fa\u9031\u8ecc\u8de1\uff0c\u6211\u5011\u5f9e\u8a13\u7df4\u968e\u6bb5\u5efa\u7acb\u7684\u3001\u4e14\u548c\u76ee\u524d\u97f3\u7bc0\u5177\u6709\u76f8\u540c\u8a9e\u5883\u985e\u578b\u4e4b\u771f\u5be6\u57fa\u9031\u8ecc\u8de1\u6536 \u96c6\u5340\u4e2d\uff0c\u53bb\u627e\u51fa\u6700\u63a5\u8fd1 GVM \u5339\u914d\u904e\u4e4b\u57fa\u9031\u8ecc\u8de1\u7684\u4e00\u500b\u771f\u5be6\u57fa\u9031\u8ecc\u8de1\u3002", "type_str": "table" }, "TABREF3": { "html": null, "num": null, "content": "
\u7528\u7684\u5269\u9918\u4e4b 60 \u500b\u8a9e\u53e5\u4f86\u4f5c\u91cf\u6e2c\u3002\u5728\u91cf\u6e2c\u5167\u90e8\u8a9e\u53e5\u7684\u5e7e\u4f55\u8ddd\u96e2\u5e73\u5747\u8aa4\u5dee\u4e4b\u5f8c\uff0c\u6211\u5011\u767c\u73fe
\u524d\u8ff0\u7684\u516d\u7a2e\u57fa\u9031\u8ecc\u8de1\u7522\u751f\u65b9\u6cd5\u4e4b\u9593\u4e26\u6c92\u6709\u660e\u986f\u7684\u6578\u503c\u5dee\u7570\uff0c\u8a73\u7d30\u7684\u5e7e\u4f55\u8ddd\u96e2\u5e73\u5747\u8aa4\u5dee\u6578
\u503c \u5982 \u8868 \u4e00 \u6240 \u793a \uff1b \u56e0 \u6b64 \u6211 \u5011 \u5c31 \u6539 \u6210 \u63a1 \u53d6 \u524d \u4eba \u63d0 \u51fa \u7684 \u539f \u7528 \u65bc \u6bd4 \u8f03 \u8f49 \u63db \u51fa \u8a9e \u97f3 (voice
conversion)
\u65b9\u6cd5MAMBMCMDMEMF
\u5e73\u5747\u8aa4\u5dee2.0662.0722.0812.0722.0752.080
\u8868\u4e00\u3001\u57fa\u9031\u8ecc\u8de1 DCT \u5411\u91cf\u4e4b\u5e7e\u4f55\u8ddd\u96e2\u5e73\u5747\u8aa4\u5dee
\u54c1\u8cea\u4e4b\u8b8a\u7570\u6578\u6bd4\u503c(VR)\u91cf\u6e2c[13]\uff0c\u4f86\u6bd4\u8f03\u9019\u516d\u7a2e\u57fa\u9031\u8ecc\u8de1\u7522\u751f\u65b9\u6cd5\uff0c\u8b8a\u7570\u6578\u6bd4\u503c\u7684\u8a08\u7b97
\u516c\u5f0f\u70ba\uff1a
11
11
", "text": "\u9019\u5169\u500b\u5340\u584a\uff0c\u6211\u5011\u6b32\u7814\u7a76\u5b83\u5011\u6240\u80fd \u767c\u63ee\u7684\u6548\u7528\uff0c\u56e0\u6b64\u6211\u5011\u63a5\u8457\u5be6\u9a57\u4e86\u516d\u7a2e\u4e0d\u540c\u7684\u57fa\u9031\u8ecc\u8de1\u7522\u751f\u65b9\u6cd5\uff0c\u9019\u4e9b\u7522\u751f\u65b9\u6cd5\u7684\u5dee\u5225 \u70ba\uff1a\u524d\u8ff0\u7684\u5169\u500b\u5340\u584a\u6709\u5426\u88ab\u5305\u542b\u9032\u53bb\uff0c\u4ee5\u53ca\u5728\u65bc\u516c\u5f0f(5)\u4e2d\u8a2d\u5b9a\u4f7f\u7528\u4e0d\u540c\u7684\u6b0a\u91cd\u503c w\u3002\u4ee5 \u4e0b\u6211\u5011\u4ee5\u7b26\u865f MA\u3001MB\u3001MC\u3001MD\u3001ME \u8207 MF \u4f86\u4ee3\u8868\u9019 6 \u7a2e\u57fa\u9031\u8ecc\u8de1\u7522\u751f\u65b9\u6cd5\uff0c\u5b83 \u5011\u7684\u7d30\u7bc0\u8a2d\u5b9a\u662f\uff1a MA\uff1a\u53ea\u4f7f\u7528 ANN \u800c\u4e0d\u4f7f\u7528 GVM \u8207 RCS\uff1b MB\uff1a\u4f7f\u7528 ANN \u548c GVM\uff0c\u4e14\u8a2d\u5b9a w=0.33\uff0c\u4f46\u4e0d\u4f7f\u7528 RCS\uff1b MC\uff1a\u4f7f\u7528 ANN \u548c GVM\uff0c\u4e14\u8a2d\u5b9a w=0.5\uff0c\u4f46\u4e0d\u4f7f\u7528 RCS\uff1b MD\uff1a\u4f7f\u7528 ANN \u548c RCS\uff0c\u4f46\u4e0d\u4f7f\u7528 GVM\uff1b ME\uff1a\u4f7f\u7528 ANN\u3001GVM \u548c RCS\uff0c\u4e14\u8a2d\u5b9a w=0.33\uff1b MF\uff1a\u4f7f\u7528 ANN\u3001GVM \u548c RCS\uff0c\u4e14\u8a2d\u5b9a w=0.5\uff1b (\u4e8c)\u3001\u5ba2\u89c0\u8a55\u4f30 \u5728\u5167\u90e8\u6e2c\u8a66\u6642\uff0c\u6211\u5011\u4ecd\u7136\u4f7f\u7528\u8a13\u7df4 ANN \u6a21\u578b\u8207 GVM \u53c3\u6578\u7684\u524d 750 \u500b\u8a9e\u53e5\uff0c\u4f86\u91cf\u6e2c\u539f \u59cb\u8a9e\u97f3\u5206\u6790\u51fa\u7684\u57fa\u9031\u8ecc\u8de1 DCT \u5411\u91cf\u548c\u7a0b\u5f0f\u7522\u751f\u7684\u57fa\u9031\u8ecc\u8de1 DCT \u5411\u91cf\u4e4b\u9593\u7684\u5e7e\u4f55\u8ddd\u96e2\u3001 \u53ca\u5169\u8005\u4e4b\u9593\u7684\u8b8a\u7570\u6578\u6bd4\u503c(variance ratio, VR)\uff1b\u800c\u5728\u5916\u90e8\u6e2c\u8a66\u6642\uff0c\u5247\u50c5\u62ff\u672a\u5728\u8a13\u7df4\u968e\u6bb5\u4f7f", "type_str": "table" }, "TABREF4": { "html": null, "num": null, "content": "
\u5982\u679c\u8981\u628a\u516d\u7a2e\u7522\u751f\u65b9\u6cd5\u5169\u5169\u4f5c\u7d44\u5408\u53bb\u4f5c\u807d\u6e2c\u5be6\u9a57\uff0c\u5247\u9700\u8981\u9032\u884c 15 \u7d44\u7684\u807d\u6e2c\u5be6\u9a57\uff0c\u5c07\u6703
\u975e\u5e38\u82b1\u8cbb\u4eba\u529b\uff0c\u56e0\u6b64\u6211\u5011\u5728\u6b64\u53ea\u9078\u64c7\u5176\u4e2d\u4e94\u7d44\u4f86\u9032\u884c\u807d\u6e2c\u5be6\u9a57\uff0c\u4e5f\u5c31\u662f(a)MA \u6bd4 MB\u3001
0.498 (b)6 \u5c0d\u97f3\u6a94\u5206\u5225\u6536\u96c6\u8a55 0.65 0.722 0.7 0.8 0.556 0.6 \u5206\uff0c\u7136\u5f8c\u5206\u5225\u8a08\u7b97\u51fa\u5e73\u5747\u8a55\u5206\u3002\u5728\u6b64\uff0c\u6211\u5011\u5c07\u5e73\u5747\u8a55\u5206\u8996\u70ba\u4e00\u7a2e\u6295\u7968\uff0c\u7576\u5e73\u5747\u8a55\u5206\u5c0f\u65bc
0.546 3 \u6642\uff0c\u5c31\u7d66\u807d\u6e2c\u6642\u5148\u64ad\u653e\u4e4b\u97f3\u6a94\u5c0d\u61c9\u7684\u7522\u751f\u65b9\u6cd5\u589e\u52a0\u4e00\u7968\uff0c\u800c\u7576\u5e73\u5747\u8a55\u5206\u5927\u65bc 3 \u6642\uff0c\u5c31 0.5
0.374 \u7d66\u807d\u6e2c\u6642\u5f8c\u64ad\u653e\u4e4b\u97f3\u6a94\u5c0d\u61c9\u7684\u7522\u751f\u65b9\u6cd5\u589e\u52a0\u4e00\u7968\u3002\u7531\u65bc\u6bcf\u500b\u7522\u751f\u65b9\u6cd5\u90fd\u6709 6 \u500b\u5408\u6210\u97f3 0.487 0.4 0.326 0.419 0.3 \u6a94\uff0c\u4e26\u4e14\u53d7\u6e2c\u8005\u5206\u6210\u5169\u7d44(\u5404 11 \u4eba)\uff0c\u6240\u4ee5\u6bcf\u4e00\u7d44\u7522\u751f\u65b9\u6cd5\u4e4b\u6bd4\u8f03\u7e3d\u5171\u6709 12 \u5f35\u7968\uff0c\u7d71\u8a08
0.249 \u6295\u7968\u7d50\u679c\u5f8c\uff0c5 \u7d44\u4f5c\u81ea\u7136\u5ea6\u807d\u6e2c\u6bd4\u8f03\u7684\u57fa\u9031\u8ecc\u8de1\u7522\u751f\u65b9\u6cd5\uff0c\u5404\u81ea\u6240\u5f97\u5230\u7684\u7968\u6578\u5c31\u5982\u5716\u516d 0.116 0.2 \u6240\u793a\u3002
0.1
00.096
insideoutside
\u5716\u4e94\u3001\u4e0d\u540c\u57fa\u9031\u8ecc\u8de1\u7522\u751f\u65b9\u6cd5\u4e4b VR \u91cf\u6e2c\u503c\u6298\u7dda
(\u4e09)\u3001\u4e3b\u89c0\u8a55\u4f30
\u4e00\u500b\u57fa\u9031\u8ecc\u8de1\u7522\u751f\u65b9\u6cd5\u7522\u751f\u51fa\u7684\u8ecc\u8de1\uff0c\u5404\u97f3\u7bc0\u7684\u8072\u8abf\u5fc5\u9808\u807d\u8d77\u4f86\u6b63\u78ba\u7121\u8aa4\uff0c\u4e26\u4e14\u8ecc\u8de1\u66f2 \u5716\u516d\u30015 \u7d44\u7522\u751f\u65b9\u6cd5\u4f5c\u807d\u6e2c\u6bd4\u8f03\u4e4b\u6295\u7968\u7d50\u679c
\u7dda\u672c\u8eab\u4e5f\u8981\u6709\u660e\u986f\u7684\u6291\u63da\u8b8a\u5316\uff0c\u5982\u6b64\u624d\u80fd\u8b93\u4eba\u807d\u8d77\u4f86\u5177\u6709\u8f03\u9ad8\u7684\u81ea\u7136\u5ea6\u3002\u5728\u6b64\u6211\u5011\u91dd\u5c0d
\u524d\u8ff0\u7684\u516d\u7a2e\u7522\u751f\u65b9\u6cd5\u9032\u884c\u4e3b\u89c0\u807d\u89ba\u4e4b\u6e2c\u8a66\uff0c\u5171\u9080\u8acb\u4e86\u5169\u7d44\u4eba\u58eb\u4f86\u53c3\u52a0\u807d\u6e2c\uff0c\u7b2c\u4e00\u7d44\u7684 \u6839\u64da\u5716\u516d\u6240\u986f\u793a\u7684\u6295\u7968\u7d50\u679c\uff0c\u6211\u5011\u53ef\u770b\u51fa MA \u6bd4
11 \u4eba\u5177\u6709\u8a9e\u97f3\u8655\u7406\u7684\u7814\u7a76\u80cc\u666f\uff0c\u7b2c\u4e8c\u7d44\u7684 11 \u4eba\u5247\u7121\u8a9e\u97f3\u8655\u7406\u4e4b\u7d93\u9a57\u3002
\u6211\u5011\u96a8\u6a5f\u9078\u53d6\u4e86\u4e09\u7bc7\u77ed\u6587\u6587\u53e5\uff0c\u7136\u5f8c\u4f7f\u7528\u524d\u8ff0\u516d\u7a2e\u7522\u751f\u65b9\u6cd5(MA \u81f3 MF)\u53bb\u5c0d\u5404\u7bc7\u6587\u53e5\u7522
\u751f\u51fa\u57fa\u9031\u8ecc\u8de1\uff0c\u63a5\u8457\u518d\u548c\u5176\u5b83\u97fb\u5f8b\u53c3\u6578(\u97f3\u7bc0\u97f3\u9577\u548c\u97f3\u91cf)\u4f5c\u7d44\u5408\uff0c\u4ee5\u5e36\u5165\u8a9e\u97f3\u4fe1\u865f\u5408\u6210
\u6a21\u7d44[14]\uff0c\u4f86\u5c0d\u5404\u7bc7\u6587\u53e5\u5408\u6210\u51fa 6 \u500b\u8a9e\u97f3\u4fe1\u865f\u6a94\uff0c\u5404\u5c0d\u61c9\u65bc\u516d\u7a2e\u57fa\u9031\u7522\u751f\u65b9\u6cd5\u4e4b\u4e00\u3002\u6b64
\u5916\uff0c\u6211\u5011\u4e5f\u628a\u7522\u751f\u51fa\u7684\u57fa\u9031\u8ecc\u8de1\u5f9e\u7537\u8072\u7684\u97f3\u9ad8\u8f49\u63db\u6210\u5973\u8072\u7684\u97f3\u9ad8\uff0c\u9019\u53ef\u900f\u904e\u8a9e\u97f3\u8f49\u63db\u9818
\u57df\u5e38\u7528\u7684\u97f3\u9ad8\u8f49\u63db\u65b9\u6cd5[10\uff0c13]\uff0c\u7136\u5f8c\u628a\u8f49\u63db\u904e\u7684\u57fa\u9031\u8ecc\u8de1\u9001\u7d66\u5148\u524d\u4ee5\u5973\u8072\u9304\u97f3\u6240\u8a13\u7df4
\u51fa\u7684 HMM \u983b\u8b5c\u6a21\u578b\uff0c\u53bb\u5408\u6210\u51fa\u5973\u8072\u57fa\u9031\u8ecc\u8de1\u7684\u97f3\u6a94\uff0c\u4ee5\u4fbf\u5c0d\u4e0d\u540c\u6027\u5225\u7684\u57fa\u9031\u8ecc\u8de1\u4f5c\u807d
\u6e2c\uff0c\u8b93\u807d\u6e2c\u5be6\u9a57\u80fd\u5920\u517c\u9867\u6027\u5225\u800c\u66f4\u5177\u6709\u4e00\u822c\u6027\u3002\u5982\u6b64\uff0c\u5c0d\u65bc\u6bcf\u4e00\u7a2e\u57fa\u9031\u8ecc\u8de1\u7522\u751f\u65b9\u6cd5\u4f86
\u8aaa\uff0c\u90fd\u6703\u6709 6 \u500b\u5408\u6210\u51fa\u7684\u8a9e\u97f3\u97f3\u6a94(3 \u7bc7\u77ed\u6587 \u00d7 2 \u7a2e\u97f3\u9ad8)\u3002
\u5728\u6b64\u807d\u6e2c\u5be6\u9a57\u7684\u9032\u884c\u65b9\u5f0f\u662f\uff0c\u6bcf\u6b21\u64ad\u653e\u5169\u500b\u5408\u6210\u8a9e\u97f3\u7684\u97f3\u6a94\u7d66\u53d7\u6e2c\u8005\u807d\uff0c\u4ee5\u6bd4\u8f03\u5169\u97f3\u6a94
\u7684\u81ea\u7136\u5ea6\uff0c\u7136\u5f8c\u8acb\u53d7\u6e2c\u8005\u6253\u4e00\u500b\u5206\u6578\uff0c\u4f86\u986f\u793a\u90a3\u4e00\u500b\u97f3\u6a94\u6bd4\u8f03\u81ea\u7136\u3002\u6253\u5206\u6578\u7684\u898f\u5247\u662f\uff0c
\u5982\u679c\u524d\u8005(\u5f8c\u8005)\u7684\u81ea\u7136\u5ea6\u660e\u986f\u9ad8\u65bc\u5f8c\u8005(\u524d\u8005)\uff0c\u5247\u7d66\u4e88 1 \u5206(5 \u5206)\uff0c\u5982\u679c\u524d\u8005(\u5f8c\u8005)\u50c5\u6bd4
\u5f8c\u8005(\u524d\u8005)\u7a0d\u597d\u4e00\u9ede\uff0c\u5247\u7d66\u4e88 2 \u5206(4 \u5206)\uff0c\u5982\u679c\u7121\u6cd5\u5340\u5206\u51fa\u5169\u8005\u7684\u81ea\u7136\u5ea6\u512a\u52a3\u5247\u7d66\u4e88 3
\u5206\u3002
", "text": "c0\u3002 \u6211\u5011\u628a\u524d\u8ff0\u516d\u7a2e\u57fa\u9031\u8ecc\u8de1\u7522\u751f\u65b9\u6cd5\u8f38\u51fa\u7684 DCT \u5411\u91cf\uff0c\u5206\u5225\u5e36\u5165\u516c\u5f0f(6)\u4f5c VR \u503c\u7684\u8a08\u7b97\uff0c \u7136\u5f8c\u628a VR \u503c\u756b\u6210\u5716\u4e94\u3002\u6839\u64da\u91cf\u6e2c\u51fa\u7684 VR \u503c\u53ef\u767c\u73fe\uff0c\u82e5\u53ea\u4f7f\u7528 ANN \u4f86\u7522\u751f\u57fa\u9031\u8ecc\u8de1 DCT \u5411\u91cf(\u5373\u65b9\u6cd5 MA)\uff0c\u5247\u91cf\u5f97\u7684 VR \u503c\u6703\u5f88\u4f4e\uff0c\u7d04\u5728 0.1 \u9644\u8fd1\u3002\u4f46\u662f\uff0c\u5982\u679c\u5728 ANN \u7522 \u751f\u51fa\u57fa\u9031\u8ecc\u8de1 DCT \u5411\u91cf\u4e4b\u5f8c\uff0c\u518d\u62ff DCT \u5411\u91cf\u53bb\u4f5c GVM \u5339\u914d(\u5373\u65b9\u6cd5 MB \u6216 MC)\u3001\u6216 RCS \u6311\u9078(\u5373\u65b9\u6cd5 MD)\uff0c\u5247\u91cf\u5f97\u7684 VR \u503c\u90fd\u6703\u6709\u986f\u8457\u7684\u63d0\u5347\uff0c\u9019\u8868\u793a\u57fa\u9031\u8ecc\u8de1 DCT \u4fc2\u6578 \u4e4b\u904e\u5ea6\u5e73\u6ed1\u73fe\u8c61\u986f\u8457\u6e1b\u5c11\uff0c\u7406\u8ad6\u4e0a\u53ef\u4ee5\u4f7f\u57fa\u9031\u8ecc\u8de1\u5f97\u5230\u66f4\u9ad8\u7684\u81ea\u7136\u5ea6\u3002\u66f4\u9032\u4e00\u6b65\uff0c\u5982\u679c \u5728 ANN \u7522\u751f\u51fa\u7684\u57fa\u9031\u8ecc\u8de1 DCT \u5411\u91cf\u4e4b\u5f8c\uff0c\u63a5\u7e8c\u4f5c GVM \u8abf\u6574\u548c RCS \u6311\u9078(\u5373\u65b9\u6cd5 ME \u6216 MF)\uff0c\u5247\u91cf\u5f97\u7684 VR \u503c\u6703\u66f4\u70ba\u63d0\u5347\u3002\u5716\u4e94\u4e2d\u7684\u5169\u689d\u66f2\u7dda\u5206\u5225\u4ee3\u8868\u62ff\u5167\u90e8\u6216\u5916\u90e8\u8a9e\u6599\u53bb \u4f5c VR \u503c\u91cf\u6e2c\u6240\u5f97\u5230\u7684\u7d50\u679c\uff0c\u7531\u9019\u5169\u689d\u66f2\u7dda\u53ef\u770b\u51fa\uff0c\u5169\u66f2\u7dda\u7684\u8b8a\u5316\u8da8\u52e2\u90fd\u8207\u524d\u9762\u8aaa\u660e\u7684 \u73fe\u8c61\u5177\u6709\u4e00\u81f4\u6027\uff0c\u56e0\u6b64\uff0c GVM \u8abf\u6574\u548c RCS \u6311\u9078\u53ef\u4ee5\u6709\u6548\u5730\u6539\u9032\u57fa\u9031\u8ecc\u8de1 DCT \u4fc2\u6578\u904e \u65bc\u5e73\u6ed1\u7684\u73fe\u8c61\uff0c\u800c\u53ef\u8b93\u57fa\u9031\u8ecc\u8de1\u7684\u81ea\u7136\u5ea6\u7372\u5f97\u63d0\u5347\u3002 MB \u6bd4 MC\u3001(c)MA \u6bd4 MC\u3001(d)MB \u6bd4 ME\u3001\u548c(e)MC \u6bd4 MF\u3002\u5c0d\u65bc\u6bcf\u4e00\u7d44\u65b9\u6cd5\u7684\u6bd4\u8f03\uff0c \u6bcf\u4e00\u500b\u53d7\u6e2c\u8005\u9808\u4f9d\u5e8f\u807d\u53d6\u5169\u500b\u7522\u751f\u65b9\u6cd5\u6240\u5408\u51fa\u7684 6 \u5c0d\u97f3\u6a94\uff0c\u4e26\u4e14\u7d66 6 \u5c0d\u97f3\u6a94\u5206\u5225\u6253\u5206 \u6578\u3002\u5728\u807d\u6e2c\u5be6\u9a57\u7d50\u675f\u4e4b\u5f8c\uff0c\u6211\u5011\u5340\u5206\u53d7\u6e2c\u8005\u6240\u96b8\u5c6c\u7684\u7d44\u5225\u3001\u4e26\u4e14\u5c0d MB \u7684\u7968\u6578\u70ba 2 \u6bd4 10\u3001MB \u6bd4 ME \u7684 \u7968\u6578\u70ba 3 \u6bd4 9\u3001\u4e26\u4e14 MC \u6bd4 MF \u7684\u7968\u6578\u70ba 4 \u6bd4 8 \u3002\u6240\u4ee5\u6211\u5011\u53ef\u8aaa\uff0c\u65b9\u6cd5 MB (ANN \u52a0 GVM)\u7522\u751f\u51fa\u7684\u57fa\u9031\u8ecc\u8de1\u8981\u6bd4\u65b9\u6cd5 MA(\u53ea\u4f7f\u7528 ANN)\u7684\u66f4\u70ba\u81ea\u7136\uff0c\u6b64\u5916\u5f9e MB \u6bd4 ME \u548c MC \u6bd4 MF \u9019\u5169\u7d44\u65b9\u6cd5\u7684\u5f97\u7968\u6578\u7d50\u679c\uff0c\u6211\u5011\u53ef\u8aaa\u4f7f\u7528 RCS (\u771f\u5be6\u8ecc\u8de1\u6311\u9078)\uff0c\u53ef\u66f4\u70ba\u63d0\u5347 \u81ea\u7136\u5ea6\u3002\u53e6\u4e00\u65b9\u9762\uff0c MB \u6bd4 MC \u7684\u7968\u6578\u70ba 6 \u6bd4 6\uff0c\u800c MA \u6bd4 MC \u7684\u7968\u6578\u70ba 7 \u6bd4 5\uff0c\u6240\u4ee5 \u5728\u81ea\u7136\u5ea6\u4e0a\uff0c\u65b9\u6cd5 MB \u548c MC \u4e4b\u9593\u4e26\u6c92\u6709\u986f\u8457\u7684\u5dee\u5225\uff0c\u4e5f\u5c31\u662f GVM \u8655\u7406\u7684\u6b0a\u91cd\u503c\u4e26\u4e0d \u6703\u9020\u6210\u660e\u986f\u7684\u5dee\u5225\u3002 \u56db\u3001\u7d50\u8ad6 \u6211\u5011\u767c\u73fe ANN \u7522\u751f\u7684\u57fa\u9031\u8ecc\u8de1 DCT \u4fc2\u6578\u5b58\u5728\u6709\u904e\u5e73\u6ed1(over smoothing)\u7684\u73fe\u8c61\uff0c\u56e0\u6b64\u5728 \u672c\u8ad6\u6587\u4e2d\uff0c\u6211\u5011\u5617\u8a66\u65bc ANN \u9810\u6e2c\u6a21\u7d44\u4e4b\u5f8c\u518d\u4e32\u63a5\u5169\u7a2e\u8655\u7406\u6a21\u7d44\uff0c\u5373 GVM \u548c RCS\uff0c\u4ee5 \u8a2d\u6cd5\u63d0\u5347 ANN \u7522\u751f\u4e4b\u57fa\u9031\u8ecc\u8de1\u7684\u81ea\u7136\u5ea6\u3002\u5c0d\u4e0d\u540c\u7522\u751f\u65b9\u6cd5\u4f5c\u5ba2\u89c0\u8a55\u4f30\u6642\uff0c\u6211\u5011\u63a1\u53d6\u4ee5 \u8a08\u7b97 VR \u503c\u4f86\u53cd\u6620\u904e\u5e73\u6ed1\u7684\u7a0b\u5ea6\uff0c\u4f9d\u64da\u91cf\u6e2c\u51fa\u7684 VR \u503c\u7d50\u679c\uff0c\u6211\u5011\u767c\u73fe GVM \u548c RCS \u6a21 \u7d44\u5169\u8005\u90fd\u80fd\u660e\u986f\u5730\u63d0\u5347 VR \u503c\uff0c\u56e0\u6b64 GVM \u548c RCS \u5169\u7a2e\u8655\u7406\u52d5\u4f5c\u78ba\u5be6\u90fd\u6709\u52a9\u65bc\u7de9\u548c\u57fa \u9031\u8ecc\u8de1 DCT \u4fc2\u6578\u4e4b\u904e\u5e73\u6ed1\u554f\u984c\uff0c\u4e26\u4e14\u7576\u628a GVM \u8207 RCS \u4e32\u63a5\u8d77\u4f86\u4f5c\u8655\u7406\u6642\uff0c\u66f4\u80fd\u5920\u9032 \u4e00\u6b65\u63d0\u5347 VR \u503c\u3002 \u53e6\u5916\u5728\u4e3b\u89c0\u8a55\u4f30\u65b9\u9762\uff0c\u6211\u5011\u9032\u884c\u4e86\u807d\u6e2c\u5be6\u9a57\uff0c\u4f86\u6bd4\u8f03\u4e94\u7d44\u57fa\u9031\u8ecc\u8de1\u7522\u751f\u65b9\u6cd5\u7684\u81ea\u7136\u5ea6\u3002 \u5728\u807d\u6e2c\u5be6\u9a57\u4e4b\u5f8c\uff0c\u628a\u53d7\u6e2c\u8005\u6240\u7d66\u7684\u8a55\u5206\u4f9d\u64da\u53d7\u6e2c\u8005\u7684\u7d44\u5225\u548c\u6240\u807d\u7684\u97f3\u6a94\uff0c\u5206\u5225\u4f5c\u6536\u96c6\u518d \u8a08\u7b97\u5e73\u5747\u8a55\u5206\uff0c\u7136\u5f8c\u628a\u5404\u500b\u5e73\u5747\u8a55\u5206\u503c\u7576\u4f5c\u5c0d\u5169\u807d\u6e2c\u97f3\u6a94\u4e4b\u81ea\u7136\u5ea6\u6bd4\u8f03\u7684\u6295\u7968\u3002\u7d71\u8a08\u7968 \u6578\u5f8c\uff0c\u6211\u5011\u767c\u73fe\u65b9\u6cd5 MB (ANN \u52a0 GVM)\u7684\u7968\u6578\u660e\u986f\u9ad8\u65bc\u65b9\u6cd5 MA (\u53ea\u4f7f\u7528 ANN)\uff1b\u6b64\u5916\uff0c \u65b9\u6cd5 ME \u7684\u7968\u6578\u9ad8\u65bc MB\uff0c\u4e14\u65b9\u6cd5 MF \u7684\u7968\u6578\u9ad8\u65bc\u65b9\u6cd5 MC\uff0c\u6240\u4ee5 RCS (\u7528\u65bc\u65b9\u6cd5 ME \u8207 MF)\u78ba\u5be6\u53ef\u6709\u6548\u5730\u63d0\u9ad8\u6240\u7522\u751f\u4e4b\u57fa\u9031\u8ecc\u8de1\u7684\u81ea\u7136\u5ea6\u3002 \u53c3\u8003\u6587\u737b", "type_str": "table" } } } }