{ "paper_id": "O14-1005", "header": { "generated_with": "S2ORC 1.0.0", "date_generated": "2023-01-19T08:04:43.053156Z" }, "title": "Automatic Multi-track Mixing by Kernel Dependency Estimation", "authors": [ { "first": "\u5433\u5b97\u5ead", "middle": [], "last": "Tsung", "suffix": "", "affiliation": { "laboratory": "", "institution": "National Central University", "location": {} }, "email": "" }, { "first": "Ting", "middle": [], "last": "Wu", "suffix": "", "affiliation": { "laboratory": "", "institution": "National Central University", "location": {} }, "email": "" }, { "first": "Chia-Hui", "middle": [], "last": "\u5f35\u5609\u60e0", "suffix": "", "affiliation": { "laboratory": "", "institution": "National Central University", "location": {} }, "email": "" }, { "first": "", "middle": [], "last": "Chang", "suffix": "", "affiliation": { "laboratory": "", "institution": "National Central University", "location": {} }, "email": "" } ], "year": "", "venue": null, "identifiers": {}, "abstract": "Due to the revolution of digital music, people can create recordings in a home studio with cheaper gear. However multi-track recordings need to be mixed to combine them into one or more channels. The question is that mixing requires background knowledge in sound engineering and psychoacoustics. It is difficult to get good mixdown for non-specialist in sound engineer. In this paper, we use supervised learning method for automatically mixing multi-track recording into coherent and well-balanced piece. Due to lack of mixing parameters, first we estimate the weight of mixing parameters by using the relation between raw multi-track and mixdown. Given the mixing parameters for any music genre, we use kernel decency estimation method to create our mixing model. The experiment show KDE is 42", "pdf_parse": { "paper_id": "O14-1005", "_pdf_hash": "", "abstract": [ { "text": "Due to the revolution of digital music, people can create recordings in a home studio with cheaper gear. However multi-track recordings need to be mixed to combine them into one or more channels. The question is that mixing requires background knowledge in sound engineering and psychoacoustics. It is difficult to get good mixdown for non-specialist in sound engineer. In this paper, we use supervised learning method for automatically mixing multi-track recording into coherent and well-balanced piece. Due to lack of mixing parameters, first we estimate the weight of mixing parameters by using the relation between raw multi-track and mixdown. Given the mixing parameters for any music genre, we use kernel decency estimation method to create our mixing model. The experiment show KDE is 42", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Abstract", "sec_num": null } ], "body_text": [], "back_matter": [], "bib_entries": { "BIBREF0": { "ref_id": "b0", "title": "Kernel Dependency Estimation", "authors": [ { "first": "J", "middle": [], "last": "Weston", "suffix": "" }, { "first": "O", "middle": [], "last": "Chapelle", "suffix": "" }, { "first": "A", "middle": [], "last": "Elisseeff", "suffix": "" }, { "first": "B", "middle": [], "last": "Scholkopf", "suffix": "" }, { "first": "V", "middle": [], "last": "Vapnik", "suffix": "" } ], "year": 2002, "venue": "Neural Information Processing Systems", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "J. Weston, O. Chapelle, A. Elisseeff, B. Scholkopf, and V. Vapnik, \"Kernel Dependency Estimation,\" Neural Information Processing Systems, 2002.", "links": null }, "BIBREF1": { "ref_id": "b1", "title": "Instrument Identification Informed Multi-track Mixing", "authors": [ { "first": "J", "middle": [], "last": "Scott", "suffix": "" }, { "first": "Y", "middle": [ "E" ], "last": "Kim", "suffix": "" } ], "year": 2013, "venue": "International Society for Music Information Retrieval", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "J. Scott and Y. E. Kim, \"Instrument Identification Informed Multi-track Mixing,\" International Society for Music Information Retrieval, 2013.", "links": null }, "BIBREF2": { "ref_id": "b2", "title": "Analysis of Acoustic Features of Automated Multi-Track Mixing", "authors": [ { "first": "J", "middle": [], "last": "Scott", "suffix": "" }, { "first": "Y", "middle": [ "E" ], "last": "Kim", "suffix": "" } ], "year": 2011, "venue": "International Society for Music Information Retrieval", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "J. Scott and Y. E. Kim, \"Analysis of Acoustic Features of Automated Multi-Track Mixing,\" International Society for Music Information Retrieval, 2011.", "links": null }, "BIBREF3": { "ref_id": "b3", "title": "Reverse Engineering of a Mix", "authors": [ { "first": "D", "middle": [], "last": "Barchiesi", "suffix": "" }, { "first": "J", "middle": [], "last": "Reiss", "suffix": "" } ], "year": 2009, "venue": "Audio Engineering Society", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "D. BARCHIESI and J. REISS, \"Reverse Engineering of a Mix,\" Audio Engineering Society, 2009.", "links": null }, "BIBREF4": { "ref_id": "b4", "title": "Automatic Target Mixing Using Least-squares Optimization of Gains And Equalization Settings", "authors": [ { "first": "D", "middle": [], "last": "Barchiesi", "suffix": "" }, { "first": "J", "middle": [], "last": "Reiss", "suffix": "" } ], "year": 2009, "venue": "", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "D. Barchiesi and J. Reiss, \"Automatic Target Mixing Using Least-squares Optimization of Gains And Equalization Settings,\" Digital Audio Effects(DAFx-09), 2009.", "links": null }, "BIBREF5": { "ref_id": "b5", "title": "Automated Production of Cross Media Content for Multi-Channel Distribution", "authors": [ { "first": "H", "middle": [], "last": "Katayose", "suffix": "" }, { "first": "A", "middle": [], "last": "Yatsui", "suffix": "" }, { "first": "M", "middle": [], "last": "Goto", "suffix": "" } ], "year": 2005, "venue": "", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "H. Katayose, A. Yatsui, and M. Goto, \"A Mix-Down Assistant Interface with Reuse of Examples,\" Automated Production of Cross Media Content for Multi-Channel Distribution, 2005.", "links": null }, "BIBREF6": { "ref_id": "b6", "title": "Mixing Secrets for the Small Studio", "authors": [ { "first": "", "middle": [], "last": "Mike", "suffix": "" }, { "first": "", "middle": [], "last": "Senior", "suffix": "" } ], "year": 2011, "venue": "", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "Mike.Senior, \"Mixing Secrets for the Small Studio,\" 2011.", "links": null }, "BIBREF7": { "ref_id": "b7", "title": "Multi-track mixing using a model of loudness and partial loudness", "authors": [ { "first": "D", "middle": [], "last": "Ward", "suffix": "" }, { "first": "J", "middle": [ "D" ], "last": "Reiss", "suffix": "" }, { "first": "C", "middle": [], "last": "Athwal", "suffix": "" } ], "year": 2012, "venue": "", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "D. Ward, J. D. Reiss, and C. Athwal, \"Multi-track mixing using a model of loudness and partial loudness,\" 2012.", "links": null }, "BIBREF8": { "ref_id": "b8", "title": "A Framework for Automatic Mixing Using Timbral Similarity Measures and Genetic Optimizatio", "authors": [ { "first": "Bennett", "middle": [], "last": "Kolasinski", "suffix": "" } ], "year": 2008, "venue": "Audio Engineering Society", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "Kolasinski and Bennett, \"A Framework for Automatic Mixing Using Timbral Similarity Measures and Genetic Optimizatio,\" Audio Engineering Society, 2008.", "links": null }, "BIBREF9": { "ref_id": "b9", "title": "Evaluation of Different Loudness Models with Music and Speech Material", "authors": [ { "first": "S", "middle": [ "H" ], "last": "Nielsen", "suffix": "" }, { "first": "E", "middle": [], "last": "Skovenborg", "suffix": "" } ], "year": 2004, "venue": "Audio Engineering Society", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "S. H. Nielsen and E. Skovenborg, \"Evaluation of Different Loudness Models with Music and Speech Material,\" Audio Engineering Society, 2004.", "links": null }, "BIBREF10": { "ref_id": "b10", "title": "Why Are Commercials so Loud? ' Perception and Modeling of the Loudness of Amplitude-Compressed Speech", "authors": [ { "first": "B", "middle": [ "C J" ], "last": "Moore", "suffix": "" }, { "first": "B", "middle": [ "R" ], "last": "Glasberg", "suffix": "" }, { "first": "M", "middle": [ "A" ], "last": "Stone", "suffix": "" } ], "year": 2003, "venue": "J. Audio Eng. Soc", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "B. C. J. Moore, B. R. Glasberg, and M. A. Stone, \"Why Are Commercials so Loud? ' Perception and Modeling of the Loudness of Amplitude-Compressed Speech,\" J. Audio Eng. Soc, 2003.", "links": null }, "BIBREF11": { "ref_id": "b11", "title": "Audio Control Facilities in Modern Recording Studios", "authors": [ { "first": "Alex", "middle": [], "last": "Balster", "suffix": "" } ], "year": 1972, "venue": "", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "Balster and Alex, \"Audio Control Facilities in Modern Recording Studios,\" Audio Engineering Society, 1972.", "links": null }, "BIBREF12": { "ref_id": "b12", "title": "Accelerating The Mixing Phase In Studio Recording productions By Automatic Audio Alignment", "authors": [ { "first": "N", "middle": [], "last": "Montecchio", "suffix": "" }, { "first": "A", "middle": [], "last": "Cont", "suffix": "" } ], "year": 2011, "venue": "International Society for Music Information Retrieval", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "N. Montecchio and A. Cont, \"Accelerating The Mixing Phase In Studio Recording productions By Automatic Audio Alignment,\" International Society for Music Information Retrieval, 2011.", "links": null }, "BIBREF13": { "ref_id": "b13", "title": "Algorithms To Measure Audio Programme Loudness And True-peak Audio Level", "authors": [ { "first": "E", "middle": [ "R R" ], "last": "", "suffix": "" } ], "year": 2010, "venue": "EBU", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "E. R. R. 128, \"Algorithms To Measure Audio Programme Loudness And True-peak Audio Level,\" EBU, 2010.", "links": null } }, "ref_entries": { "FIGREF0": { "type_str": "figure", "num": null, "text": "Production)\u4e0a\u5927\u81f4\u5206\u70ba\u4e09\u500b\u968e\u6bb5\uff0c\u5275\u4f5c\u7de8\u66f2(Pre-Production)\u3001 \u8072\u97f3\u9304\u88fd(Production)\u3001\u5f8c\u88fd(Post-Production)\u3002\u5176\u4e2d\u5f8c\u88fd\u53c8\u5206\u70ba\u6df7\u97f3(Mixing)\u53ca\u6bcd\u5e36\u5f8c\u88fd (Mastering)\u5169\u90e8\u5206; \u6df7\u97f3\u5728\u97f3\u6a02\u88fd\u4f5c\u4e0a\u662f\u500b\u975e\u5e38\u91cd\u8981\u7684\u904e\u7a0b; \u5176\u4e3b\u8981\u7684\u5de5\u4f5c\u662f\u8981\u628a\u5148\u524d \u9304\u88fd\u597d\u7684\u591a\u8ecc(Multi-Track)\u7684\u8072\u97f3\uff0c\u5982\u4eba\u8072\u3001\u5409\u4ed6\u3001\u7235\u58eb\u9f13\u7b49\u8072\u8ecc\u6df7\u5408\u9032\u540c\u4e00\u500b\u7acb\u9ad4\u8072 \u8ecc(stereo channel)\u6216\u55ae\u8072\u8ecc(Mono Channel)\u4e2d\u3002 \u8fd1\u5e74\u4f86\u7531\u65bc\u6578\u4f4d\u97f3\u6a02\u7684\u84ec\u52c3\u767c\u5c55\uff0c\u9304\u97f3\u5668\u6750\u8d8a\u4f86\u8d8a\u666e\u53ca\u3002\u4f7f\u5f97\u975e\u6df7\u97f3\u5c08\u696d\u4eba\u58eb\u4e5f\u80fd \u5229\u7528\u9304\u97f3\u754c\u9762(Audio Interface)\u9304\u88fd\u51fa\u4e0d\u932f\u7684\u6210\u54c1; \u4f46\u662f\u4e00\u65e6\u9304\u88fd\u4e86\u591a\u8ecc(Multi-Track Recording)\u5c31\u6703\u9762\u81e8\u5230\u6df7\u97f3(Mixing)\u7684\u554f\u984c\uff0c\u5373\u9700\u8981\u628a\u591a\u8ecc\u7684\u8072\u97f3\u6df7\u5408\u5728\u540c\u4e00\u500b\u8ecc\u4e2d; \u6df7 \u97f3\u727d\u626f\u5230\u8a31\u591a\u97f3\u97ff\u53ca\u8072\u5b78\u5fc3\u7406\u5b78\u7684\u76f8\u95dc\u6280\u8853\u8207\u77e5\u8b58\uff0c\u975e\u5c08\u696d\u4eba\u58eb\u8981\u6df7\u51fa\u5c1a\u53ef\u7684\u6210\u54c1\u6709\u4e00 \u5b9a\u7684\u96e3\u5ea6\uff0c\u6df7\u51fa\u4f86\u7684\u7d50\u679c\u5f80\u5f80\u6703\u7167\u6210\u6574\u9996\u6b4c\u8046\u807d\u7684\u6e05\u6670\u5ea6\u964d\u4f4e\u3001\u8072\u97f3\u4e0d\u624e\u5be6\u3001\u97f3\u91cf\u843d\u5dee \u592a\u5927\u3001\u7a7a\u9593\u611f\u4e0d\u5920\u3001\u8072\u97f3\u96dc\u4e82\u7b49\u554f\u984c; \u800c\u4e14\u6df7\u97f3\u7684\u8655\u7406\u65b9\u5f0f\u57fa\u672c\u4e0a\u6703\u96a8\u8457\u6a02\u5668\u3001\u97f3\u6a02\u985e \u578b\u800c\u6709\u6240\u4e0d\u540c\uff0c\u4e0d\u540c\u7684\u97f3\u6a02\u985e\u578b\u6703\u6709\u4e0d\u540c\u7684\u6df7\u97f3\u98a8\u683c(Mixing Style)\uff0c\u9019\u66f4\u52a0\u589e\u52a0\u4e86\u4e00\u822c \u975e\u5c08\u696d\u4eba\u58eb\u5b78\u7fd2\u6df7\u97f3\u7684\u96e3\u5ea6\u3002\u6240\u4ee5\u8981\u5982\u4f55\u85c9\u7531\u96fb\u8166\u4f86\u5e6b\u52a9\u6df7\u97f3\u4fbf\u662f\u672c\u7bc7\u8ad6\u6587\u7684\u76ee\u6a19\u3002\u63a5 \u4e0b\u4f86\u5c07\u6703\u8a73\u7d30\u4ecb\u7d39\u6df7\u97f3\u7684\u76f8\u95dc\u80cc\u666f\u77e5\u8b58\u3002", "uris": null }, "TABREF0": { "num": null, "type_str": "table", "html": null, "content": "
1.1 \u6df7\u97f3(Mixing) \u6df7\u97f3\u5728\u97f3\u6a02\u88fd\u4f5c\u4e0a\u662f\u975e\u5e38\u91cd\u8981\u7684\u904e\u7a0b\uff0c\u4e0d\u540c\u7684\u6df7\u97f3\u65b9\u5f0f\u5728\u6700\u5f8c\u7684\u6210\u54c1\u4e0a\u6703\u6709\u622a\u7136\u4e0d \u540c\u7684\u6a21\u6a23\u3002\u6df7\u97f3\u7684\u597d\u58de\u6703\u5f71\u97ff\u6574\u9996\u6b4c\u7684\u8868\u73fe\uff0c\u597d\u7684\u6df7\u97f3\u53ef\u4ee5\u63a9\u84cb\u7455\u75b5\u3001\u653e\u5927\u512a\u9ede\uff0c\u63d0\u5347 \u6574\u9ad4\u7684\u8cea\u611f\u3002\u5728\u6df7\u97f3\u7684\u904e\u7a0b\u4e2d\uff0c\u6df7\u97f3\u5e2b(Mixing Engineer)\u6703\u4f9d\u7167\u5404\u97f3\u8ecc/\u6a02\u5668\u9593\u7684\u983b\u7387 (Frequency) \u3001\u97ff\u5ea6(loudness) \u3001\u97f3\u8272\u3001\u97f3\u5834\u5b9a\u4f4d(Panoramic Position) \u3001\u7a7a\u9593\u611f\u7b49\u8072\u97f3\u5143 \u7d20\u52a0\u4ee5\u8abf\u914d\uff0c\u4ee5\u8b93\u6bcf\u500b\u97f3\u8ecc(track)/\u6a02\u5668\u6700\u4f73\u5316\uff0c\u8b93\u6bcf\u500b\u97f3\u8ecc\u5728\u6700\u5f8c\u6df7\u5728\u4e00\u8d77\u6642\u4e00\u6a23\u80fd\u4fdd \u6301\u6e05\u6670\uff0c\u4fdd\u6709\u5c64\u6b21\uff0c\u4f7f\u5f97\u97f3\u6a02\u5448\u73fe\u66f4\u751f\u52d5\u3001\u66f4\u52d5\u807d\u3002 \u5728\u958b\u59cb\u6df7\u97f3\u524d\uff0c\u6df7\u97f3\u5e2b\u6703\u5148\u4f5c\u6df7\u97f3\u898f\u5283(Mixing Design)\uff0c\u898f\u5283\u6574\u9996\u6b4c\u7684\u97f3\u50cf(Sound Image)\uff0c\u6c7a\u5b9a\u6bcf\u500b\u97f3\u8ecc\u5728\u6574\u9996\u6b4c\u88e1\u7684\u5b9a\u4f4d\u4ee5\u53ca\u5176\u91cd\u8981\u6027\u3002\u5982\u4e0b\u5716 1 \u70ba\u4e00\u9996\u7235\u58eb\u6a02\u7684\u6df7\u97f3 \u898f\u5283\u793a\u610f\u5716\uff0c\u6211\u5011\u53ef\u4ee5\u767c\u73fe\u5728\u9019\u9996\u6b4c\u4e2d\u4eba\u8072(Vocal)\u8a2d\u8a08\u5728\u97f3\u50cf\u7684\u6b63\u4e2d\u9593\uff0c\u5409\u4ed6(Guitar) \u5206\u5225\u843d\u5728\u4eba\u8072\u7684\u5de6\u53f3; \u97f3\u91cf\u65b9\u9762\u4e3b\u5409\u4ed6(Lead Guitar)\u7565\u5927\u65bc\u4eba\u8072\u7b49\u7b49\uff0c\u6df7\u97f3\u7684\u904e\u7a0b\u4e2d\u4e3b \u8981\u5c31\u662f\u8abf\u914d\u9019\u4e9b\u97f3\u91cf(Volume)\u3001\u7b49\u5316(Equalization)\u3001\u64fa\u4f4d(Pan)\u4f86\u9054\u6210\u6211\u5011\u7684\u6df7\u97f3\u898f\u5283\u8b93 \u6574\u9ad4\u66f4\u52a0\u548c\u8ae7\uff0c\u8b93\u500b\u6a02\u5668\u878d\u5165\u5176\u4e2d\u3002 \u5716 1 \u6df7\u97f3\u8a2d\u8a08\u793a\u610f\u5716\uff1a\u4eba\u8072(Vocal)\u5c45\u4e2d \u5716\u7247\u4f86\u6e90\uff1aThe Art of Mixing -David Gibson 1.2 \u7814\u7a76\u52d5\u6a5f(Motivation) \u5728\u81ea\u52d5\u6df7\u97f3(Automatic Mixing)\u7684\u7814\u7a76\u4e2d\uff0c\u5927\u591a\u90fd\u662f\u5229\u7528\u8072\u97f3\u7279\u5fb5\u9593\u7684\u95dc\u4fc2\u4f30\u8a08\u4ee5\u53ca \u9810\u6e2c\u5176\u6df7\u97f3\u7684\u53c3\u6578\uff0c\u6240\u4f7f\u7528\u7684\u8072\u97f3\u9577\u5ea6\u5927\u591a\u70ba\u4e00\u9996\u6b4c\u4e2d 30 \u79d2\u7684\u7247\u6bb5\uff0c\u6700\u5f8c\u6240\u6df7\u97f3\u51fa\u4f86 \u7684\u6210\u54c1\u7576\u7136\u4e5f\u6703\u76f8\u4f3c\u65bc\u9019 30 \u79d2\u7684\u7247\u6bb5\u3002\u4f46\u662f\u6df7\u97f3\u5be6\u969b\u4e0a\u662f\u6703\u56e0\u70ba\u6a4b\u6bb5\u7684\u4e0d\u540c\u800c\u6709\u4e0d\u540c \u7684\u6df7\u97f3\u65b9\u5f0f\u7684\uff0c\u4f8b\u5982\u5728\u4e3b\u6b4c(Verse)\u8207\u526f\u6b4c(Chorus)\u7684\u6df7\u97f3\u65b9\u5f0f\u6703\u662f\u4e0d\u4e00\u6a23\u7684\uff0c\u5f8c\u8005\u5176\u6a4b \u6bb5\u901a\u5e38\u70ba\u6b4c\u66f2\u4e2d\u6fc0\u6602\u7684\u90e8\u4efd\uff0c\u914d\u5668\u4f7f\u7528\u6703\u524d\u8005\u591a\uff0c\u5728\u97f3\u91cf\u6216\u983b\u7387\u4e0a\u6bd4\u4f8b\u6703\u6709\u6240\u4e0d\u540c\u3002 \u518d\u8005\u6b4c\u66f2\u97f3\u6a02\u985e\u578b\u4e5f\u6703\u5f71\u97ff\u6df7\u97f3\u7684\u65b9\u5f0f\uff0c\u540c\u6a23\u662f\u7235\u58eb\u9f13\u5728\u6d41\u884c\u6b4c\u8207\u7235\u58eb\u6a02\u4e2d\u97f3\u8272\u8207 \u5176\u5360\u6709\u7684\u6bd4\u4f8b\u4e5f\u6703\u6709\u6240\u4e0d\u540c\uff0c\u540c\u6a23\u7684\u6a02\u5668\u5728\u4e0d\u540c\u7684\u97f3\u6a02\u985e\u578b\u4e2d\u6703\u6709\u4e0d\u540c\u7684\u97f3\u8272\u4ee5\u53ca\u89d2\u8272\uff0c \u4f8b\u5982\uff0c\u5728\u7235\u58eb\u6a02\u4e2d\u9f13\u624b\u4e5f\u53ef\u4ee5\u662f\u6a02\u66f2\u4e2d\u7684\u4e3b\u89d2\uff0c\u5728\u6d41\u884c\u6b4c\u4e2d\u4e3b\u89d2\u5f80\u5f80\u662f\u4eba\u8072\u6216\u662f\u96fb\u5409\u4ed6\u3002 \u6240\u4ee5\u82e5\u5229\u7528 30 \u79d2\u7247\u6bb5\u5efa\u7acb\u7684\u6a21\u578b\u4f86\u5957\u7528\u5728\u6574\u9996\u6b4c\u66f2\u7684\u6df7\u97f3\u5c07\u6703\u9020\u6210\u6574\u9996\u6b4c\u8f03\u5e73\u6de1\u7121\u5473; \u5728 Jeffrey Scott et al.[2]\u7684\u8a2a\u554f\u4e2d\u4e5f\u6709\u63d0\u5230\uff0c\u6df7\u97f3\u5176\u5be6\u662f\u500b\u5225\u7684(Case By Case)\uff0c\u6df7\u97f3\u5e2b\u57fa \u672c\u4e0a\u90fd\u6703\u4f9d\u7167\u5404\u97f3\u8ecc\u7684\u8072\u97ff\u3001\u97f3\u6a02\u985e\u578b\u4e0d\u540c\u800c\u6709\u4e0d\u540c\u7684\u8655\u7406\u65b9\u5f0f\uff0c\u518d\u8005\u97f3\u6a02\u611f\u53d7\u662f\u975e\u5e38 \u4e3b\u89c0\u7684\uff0c\u4e0d\u540c\u7684\u4eba\u6703\u6709\u4e0d\u540c\u7684\u504f\u597d\uff0c\u6240\u4ee5\u6bd4\u8f03\u96e3\u53bb\u8a02\u51fa\u4e00\u500b\u901a\u5247(General Rule)\u4f86\u9032\u884c\u6df7 \u97f3\u3002 \u4e00\u500b\u6df7\u97f3\u6a21\u578b\u5305\u62ec\u97f3\u91cf\u3001\u983b\u7387(Equalization)\u3001\u6a02\u5668\u64fa\u4f4d(Panning)\u7b49\u591a\u500b\u53c3\u6578\uff0c\u5927\u591a \u6578\u7684\u7814\u7a76\u90fd\u662f\u7368\u7acb\u7684\u53bb\u9810\u6e2c\u5404\u8ecc\u7684\u6df7\u97f3\u53c3\u6578\uff0c\u5982\u591a\u7dda\u6027\u8ff4\u6b78\uff0c\u5229\u7528\u5404\u8ecc\u7279\u5fb5\u9593\u7684\u95dc\u4fc2\u53bb \u5efa\u7acb\u5404\u8ecc\u7684\u8ff4\u6b78\u6a21\u578b\u3002\u4f46\u5be6\u969b\u4e0a\u6bcf\u8ecc\u9593\u7684\u6df7\u97f3\u53c3\u6578\u662f\u6709\u4f9d\u8cf4\u95dc\u4fc2\u7684(Dependency)\uff0c\u8209\u4f8b \u4f86\u8aaa\u6709\u4e00\u8ecc\u7684\u97f3\u91cf\u4e0a\u5347\u52e2\u5fc5\u5c31\u6703\u6709\u4e00\u8ecc\u7684\u97f3\u91cf\u4e0b\u964d\uff0c\u9019\u6a23\u6574\u9996\u6b4c\u7684\u97f3\u91cf\u624d\u4e0d\u6703\u5ffd\u5927\u5ffd\u5c0f\uff0c \u82e5\u662f\u5c0d\u5404\u8ecc\u7368\u7acb\u53bb\u5efa\u7acb\u5176\u6a21\u578b\u7684\u8a71\uff0c\u6700\u5f8c\u7684\u6210\u54c1\u5c07\u6703\u55aa\u5931\u5176\u4f9d\u8cf4\u95dc\u4fc2\u3002\u6240\u4ee5\u5728\u53c3\u6578\u9810\u6e2c \u7684\u65b9\u6cd5\u9078\u7528\u4e0a\u6211\u5011\u8a8d\u70ba\u9700\u8981\u8003\u616e\u5230\u4f9d\u8cf4\u6027\u7684\u554f\u984c\u3002 \u7531\u4ee5\u4e0a\u4e09\u9ede\uff0c\u672c\u7bc7\u8ad6\u6587\u63a1\u7528\u5c0d\u4e0d\u540c\u7684\u97f3\u6a02\u985e\u578b\u4e0d\u540c\u7684\u6a4b\u6bb5\uff0c\u5229\u7528\u548c\u4f9d\u8cf4\u4f30\u8a08(Kernel Dependency Estimation)[1]\u7684\u65b9\u6cd5\u4f86\u5efa\u7acb\u6700\u5f8c\u7684\u6a21\u578b\u3002\u9996\u5148\u6211\u5011\u6703\u5148\u9700\u8981\u4f7f\u7528\u8005\u5148\u63d0\u4f9b\u4e00 \u4e9b\u8a72\u6b4c\u66f2\u7684\u8a0a\u606f\uff0c\u4f8b\u5982\u97f3\u6a02\u5f62\u614b\u3001\u6a4b\u6bb5\u3001\u5206\u8ecc\u7684\u6a02\u5668\u6a19\u7c64\u7b49\u7b49\uff0c\u63a5\u8457\u5728\u5c0d\u500b\u5225\u7684\u6a4b\u6bb5\u5957 \u7528\u8a72\u97f3\u6a02\u985e\u578b\u7684\u6838\u4f9d\u8cf4\u6a21\u578b\u9810\u6e2c\u51fa\u6df7\u97f3\u53c3\u6578\u5f8c\u5373\u5b8c\u6210\u6df7\u97f3\u3002(\u5716 2) \u5716 2 \u7cfb\u7d71\u4f7f\u7528\u6982\u5ff5\u5716 \u672c\u7bc7\u8ad6\u6587\u7684\u67b6\u69cb\u5982\u4e0b\uff0c\u7b2c\u4e8c\u7ae0\u5c07\u6703\u8a0e\u8ad6\u81ea\u52d5\u6df7\u97f3\u7684\u76f8\u95dc\u7814\u7a76\uff0c\u7b2c\u4e09\u7ae0\u5247\u6703\u4ecb\u7d39\u672c\u7bc7 \u8ad6\u6587\u7684\u7814\u7a76\u65b9\u6cd5\u53ca\u6240\u4f7f\u7528\u7684\u8cc7\u6599\u96c6\uff0c\u7b2c\u56db\u7ae0\u70ba\u5be6\u9a57\uff0c\u6211\u5011\u6703\u5c0d\u6211\u5011\u7684\u6df7\u97f3\u6a21\u578b\u53bb\u505a\u4ea4\u53c9 \u9a57(Cross validation) \u4f86\u8a55\u4f30\u6a21\u578b\u7684\u6b63\u78ba\u6027\u4ee5\u53ca\u4f9d\u8cf4\u6027\u7684\u529f\u7528\uff0c\u7b2c\u4e94\u7ae0\u70ba\u7d50\u8ad6\u4ee5\u53ca\u672a\u4f86\u5de5 \u4f5c\u3002 \u4e8c\u3001 \u76f8\u95dc\u7814\u7a76 \u5728\u4ecb\u7d39\u6df7\u97f3\u53c3\u6578\u9810\u6e2c\u6a21\u578b\u4e4b\u524d\uff0c\u6211\u5011\u5fc5\u9808\u8a2d\u6cd5\u53d6\u5f97\u4e00\u4e9b\u6b4c\u66f2\u7684\u6df7\u97f3\u53c3\u6578\u4ee5\u4f5c\u70ba\u8a13\u7df4 \u8cc7\u6599\uff0c\u4e0d\u904e\u6df7\u97f3\u53c3\u6578\u5728\u5be6\u969b\u4e0a\u662f\u96e3\u4ee5\u53d6\u5f97\u7684\uff0c\u7531\u65bc\u6df7\u97f3\u5e2b\u4f7f\u7528\u7684\u8edf\u9ad4\u4e0d\u540c\uff0c\u4e0d\u540c\u7684\u5668\u6750 /\u8edf\u9ad4\u6703\u6709\u4e0d\u540c\u7684\u8a2d\u5b9a\u3001\u57fa\u6e96\u53ca\u523b\u5ea6\uff0c\u9019\u8b93\u5f97\u5230\u6df7\u97f3\u53c3\u6578\u9019\u4e00\u985e\u7684\u8cc7\u8a0a\u8b8a\u5f97\u975e\u5e38\u56f0\u96e3\uff0c \u4e14\u6df7\u97f3\u5e2b\u5728\u6df7\u97f3\u6642\u4e5f\u9bae\u5c11\u6703\u628a\u53c3\u6578\u8a18\u9304\u4e0b\u4f86\u3002\u6240\u4ee5\u5728\u591a\u8ecc\u6df7\u97f3(Multitrack Mixing)\u7684\u76f8\u95dc \u7814\u7a76\u4e2d\u5927\u90e8\u4efd\u7684\u7814\u7a76\u8457\u91cd\u65bc\u5982\u4f55\u4f30\u8a08\u51fa\u6df7\u97f3\u53c3\u6578\uff0c\u5305\u62ec\uff1a\u97f3\u91cf(Volume)\u3001\u983b\u7387(Frequency)\u3001 \u52d5\u614b(Dynamic)\u7b49\u7b49\u3002\u53e6\u4e00\u90e8\u4efd\u7684\u7814\u7a76\u8457\u91cd\u5728\u5982\u4f55\u5efa\u7acb\u6df7\u97f3\u53c3\u6578\u6a21\u578b\u3002 2.1. \u6df7\u97f3\u57fa\u672c\u5143\u7d20(Basic Factor of Mixing) \u5728\u6df7\u97f3\u7684\u904e\u7a0b\u4e2d\u97f3\u91cf\u5e73\u8861\u662f\u4ef6\u5f88\u91cd\u8981\u7684\u4e8b\uff0c\u6df7\u97f3\u5e2b(Mixing Engineer)\u6703\u8abf\u6574\u6bcf\u500b\u97f3 \u8ecc\u9593\u5f7c\u6b64\u4e4b\u9593\u7684\u97f3\u91cf(\u5716 1)\uff0c\u6c7a\u5b9a\u5404\u8ecc\u5728\u9019\u9996\u6b4c\u4e2d\u7684\u97f3\u91cf\u6bd4\u4f8b\uff0c\u5373\u662f\u6c7a\u5b9a\u5404\u8ecc\u5728\u97f3\u50cf (Image)\u7684\u524d\u5f8c\u9806\u5e8f; \u82e5\u5176\u4e2d\u4e00\u97f3\u8ecc\u7684\u97f3\u91cf\u6bd4\u5176\u4ed6\u97f3\u8ecc\u9084\u8981\u5927\u5f88\u591a\u7684\u8a71\uff0c\u6574\u9996\u6b4c\u5c07\u6703\u807d\u8d77 \u4f86\u982d\u91cd\u8173\u8f15\u3002 \u5728\u97f3\u8272\u4fee\u6b63\u7684\u904e\u7a0b\u4e2d\uff0c\u6df7\u97f3\u5e2b(Mixing Engineer)\u6703\u7528\u5230\u7b49\u5316\u5668(Equalizer, EQ)\u7684\u5de5\u5177 \u4f86\u53bb\u5c0d\u6bcf\u500b\u97f3\u8ecc\u7684\u983b\u7387\u505a\u4fee\u6b63\u8abf\u6574\u3002\u4f8b\u5982\u6211\u5011\u767c\u73fe\u5409\u4ed6\u7684\u97f3\u8272\u8ddf\u5176\u4ed6\u7684\u6a02\u5668\u76f8\u6bd4\u592a\u4eae\u592a \u5c16\u92b3\uff0c\u4ee5\u81f3\u65bc\u7121\u6cd5\u878d\u5165\u9019\u9996\u6b4c\u4e2d\uff0c\u6211\u5011\u5c31\u53ef\u4ee5\u7528\u7b49\u5316\u5668\u53bb\u5c0d\u5409\u4ed6\u7684\u9ad8\u983b\u90e8\u5206\u505a\u8870\u6e1b(cut)\u3002 \u983b\u7387\u9593\u7684\u5e73\u8861\u5728\u6df7\u97f3\u904e\u7a0b\u4e2d\u4e5f\u662f\u91cd\u8981\u7684\u904e\u7a0b\u4e4b\u4e00\u3002 \u6a02\u5668\u64fa\u4f4d\u662f\u5c07\u9304\u88fd\u597d\u7684\u8072\u97f3\u8a0a\u865f\u653e\u7f6e\u65bc\u65b0\u7684\u96d9\u8072\u9053\u6216\u591a\u8072\u9053\u7684\u8072\u5834(Sound Field)\u3002 \u7531\u65bc\u6211\u5011\u4e00\u822c\u97f3\u97ff\u8a2d\u5099\u7684\u74b0\u5883\u57fa\u672c\u4e0a\u662f\u4ee5\u96d9\u8072\u9053\u70ba\u4e3b(stereo)\uff0c\u96d9\u8072\u9053\u7684\u6df7\u97f3\u53ef\u4ee5\u5728\u8046\u807d \u4e0a\u589e\u52a0\u5e73\u9762\u7684\u807d\u611f\uff0c\u800c\u4e0d\u662f\u53ea\u6709\u4e00\u9ede;\u6240\u4ee5\u6211\u5011\u5728\u6df7\u97f3\u7684\u6642\u5019\u6703\u6c7a\u5b9a\u5404\u6a02\u5668\u7684\u64fa\u4f4d\uff0c\u770b \u662f\u8981\u64fa\u5728\u4e2d\u9593\u9084\u662f\u64fa\u5728\u9760\u8fd1\u5de6\u8072\u9053/\u5de6\u5587\u53ed\u9084\u662f\u53f3\u8072\u9053/\u53f3\u5587\u53ed\u7b49\u7b49\u4f86\u589e\u52a0\u6574\u9ad4\u7684\u7a7a\u9593\u611f\uff0c Left_output = cos(p)*input Right_output = sin(p)*input (1) \u5176\u4e2d p \u70ba\u504f\u96e2\u4e2d\u592e\u9ede\u7684\u89d2\u5ea6\uff0cLeft_output, Right_output \u5206\u5225\u70ba\u5de6\u8072\u9053\u53f3\u8072\u9053\u7684\u8f38\u51fa\u3002 2.2. \u6df7\u97f3\u53c3\u6578\u4f30\u8a08(Mixing Parameter Estimation) \u5982\u524d\u4e00\u7bc0\u7814\u7a76\u52d5\u6a5f\u6240\u63d0\u5230\uff0c\u7531\u65bc\u6df7\u97f3\u53c3\u6578\u96e3\u4ee5\u53d6\u5f97\uff0c\u6211\u5011\u9700\u8981\u5229\u7528\u539f\u59cb\u5206\u8ecc\u548c\u6df7\u97f3 \u6210\u54c1(Mixdown)\u9593\u7684\u95dc\u4fc2\u4f86\u4f30\u8a08\u51fa\u6bcf\u9996\u6b4c\u7684\u6df7\u97f3\u53c3\u6578\u3002\u5728 Jeffrey Scott et al. [2]\u7684\u7814\u7a76\u4e2d\uff0c \u4ed6\u5011\u63a1\u7528\u4e86\u8a2a\u554f(interview)\u7684\u65b9\u5f0f\uff0c\u8a2a\u554f\u4e86\u7dda\u4e0a\u7684\u6df7\u97f3\u5e2b(Mixing Engineer)\u4e86\u89e3\u4ed6\u5011\u5982\u4f55 \u6df7\u97f3\u3001\u5982\u4f55\u8655\u7406\u8072\u97f3\uff0c\u5229\u7528\u8a2a\u554f\u5f8c\u5f97\u5230\u7684\u4e00\u4e9b\u901a\u5247\u4f86\u7576\u4f5c\u4ed6\u5011\u6700\u5f8c\u6df7\u97f3\u6a21\u578b\u5efa\u7acb\u7684\u4f9d\u64da\u3002 \u5728\u5927\u90e8\u5206\u81ea\u52d5\u6df7\u97f3(Automatic Mixing)\u7684\u7814\u7a76\u4e2d[3, 4]\uff0c\u5728\u6df7\u97f3\u53c3\u6578\u4f30\u8a08\u4e0a\u5927\u591a\u662f\u5047\u8a2d\u5176 \u8a08\u3002 \u5716 3 \u6700\u5c0f\u5e73\u65b9\u6cd5 \u7531\u65bc\u63a1\u7528\u7dda\u6027\u7d44\u5408\u7684\u5047\u8a2d\uff0c\u6bcf\u8ecc\u9593\u5f7c\u6b64\u8981\u662f\u7dda\u6027\u7368\u7acb(Linearly independent)\u7684\uff0c\u9019\u6703 \u5f71\u97ff\u6700\u5f8c\u8072\u97f3\u7279\u5fb5\u7684\u9078\u7528 \u4ee5\u53ca\u97f3\u8ecc\u7684\u9078\u64c7\uff0c\u9078\u64c7\u542b\u6709\u8f03\u591a\u4e32\u97f3\u7684\u97f3\u8ecc (\u4f8b\u5982\u9304\u88fd overhead \u6642\u6703\u9023\u5927\u9f13\u5c0f\u9f13\u7b49\u5176\u4ed6\u9f13\u7d44\u7684\u8072\u97f3\u4e00\u4f75\u9304\u9032)\uff0c\u5728\u5be6\u52d9\u4e0a\u56e0\u70ba\u5305\u542b\u591a\u500b\u6a02\u5668\u7684 \u8072\u97f3\uff0c\u4f7f\u5f97\u97f3\u8ecc\u9593\u5f7c\u6b64\u4e26\u975e\u7368\u7acb\u6703\u9020\u6210\u5176\u4f30\u8a08\u7d50\u679c\u6703\u6709\u8aa4\u5dee\u3002 2.3. \u6df7\u97f3\u53c3\u6578\u9810\u6e2c \u4e09\u3001 \u7814\u7a76\u65b9\u6cd5 \u672c\u7cfb\u7d71\u4e3b\u8981\u6d41\u7a0b\u5982\u4e0b\u5716 4 \u6240\u793a\uff0c\u6211\u5011\u6703\u5148\u628a\u539f\u59cb\u7684\u5206\u8ecc\u9304\u97f3\u6a94\u4f9d\u7167\u4f7f\u7528\u8005\u63d0\u4f9b\u8a72 \u97f3\u6a02\u7684\u8cc7\u8a0a\u505a\u524d\u8655\u7406\uff0c\u63a5\u4e0b\u4f86\u505a\u8072\u97f3\u7279\u5fb5\u7684\u64f7\u53d6(Feature Extraction)\u4ee5\u4fbf\u4e4b\u5f8c\u6a21\u578b(Model) \u7684\u8a13\u7df4\u53ca\u6e2c\u8a66\u3002\u7531\u65bc\u4e0d\u540c\u7684\u985e\u578b\u3001\u4e0d\u540c\u6a4b\u6bb5\u7684\u97f3\u6a02\u6703\u6709\u4e0d\u540c\u7684\u6df7\u97f3\u65b9\u5f0f\uff0c\u5728\u6a21\u578b\u5efa\u7acb\u6642 \u6211\u5011\u6703\u7279\u5225\u4f9d\u7167\u4e0d\u540c\u7684\u97f3\u6a02\u985e\u578b\u5efa\u7acb\u500b\u5225\u7684\u6df7\u97f3\u53c3\u6578\u6a21\u578b\uff0c\u5728\u4f9d\u4e0d\u540c\u7684\u6a4b\u6bb5\u53bb\u5efa\u7acb\u6a21\u578b\uff0c \u5982\u5716 5 \u6211\u5011\u6703\u5c0d ROCK \u7684\u7684\u97f3\u6a02\u985e\u578b\u5efa\u7acb Intro\u3001Verse\u3001Chorus\uff0cPOP \u97f3\u6a02\u985e\u578b\u4e5f\u5efa\u7acb \u5176\u4e09\u500b\u6a4b\u6bb5\u7684\u6a21\u578b\uff0c\u4ee5\u6b64\u985e\u63a8\u3002\u8a13\u7df4(Training)\u7684\u90e8\u5206\u6709\u5169\u5927\u6b65\u9a5f; \u7b2c\u4e00\u6b65\u9a5f\u662f\u6df7\u97f3\u53c3\u6578 \u4f30\u8a08(Parameter Estimation)\uff0c\u7531\u65bc\u6df7\u97f3\u53c3\u6578\u96e3\u4ee5\u53d6\u5f97\uff0c\u539f\u59cb\u8a13\u7df4\u8cc7\u6599\u4e2d\u4e5f\u7121\u6b64\u8cc7\u8a0a\uff0c\u6211 \u5011 \u6703 \u5148 \u5229 \u7528 \u539f \u59cb \u5206 \u8ecc \u9304 \u97f3 \u548c \u6df7 \u97f3 \u6210 \u54c1 (Mixdown) \u505a \u6700 \u5c0f \u5e73 \u65b9 \u6cd5 \u4f30 \u8a08 (Least Square Estimation)\uff0c\u85c9\u6b64\u4f86\u4f30\u8a08\u51fa\u8a13\u7df4\u8cc7\u6599\u4e2d\u6bcf\u9996\u6b4c\u7684\u6df7\u97f3\u53c3\u6578\u7684\u6b0a\u91cd(Weight of Mixing Parameter) \u3002 3.1 \u8cc7\u6599\u96c6 (DATASET) \u6211\u5011\u4f7f\u7528\u7684\u8cc7\u6599\u96c6(Data Set)\u662f\u4f86\u81ea\u570b\u5916\u4e00\u672c\u95dc\u65bc\u6df7\u97f3\u7684\u5c08\u9580\u66f8\u7c4d\"Mixing Secrets for Small Studio\"[7]\uff0c\u6b64\u66f8\u6709\u63d0\u4f9b\u591a\u9996\u539f\u59cb\u5206\u8ecc\u6a94\u6848\u7d66\u8b80\u8005\u7528\u65bc\u6df7\u97f3\u7df4\u7fd2\u7528\uff0c\u5176\u542b\u62ec\u7684 \u97f3\u6a02\u985e\u578b\u6416\u6efe\u3001\u7235\u58eb\u3001\u9109\u6751\u7b49\u591a\u7a2e\u97f3\u6a02\u985e\u578b\uff0c\u5982\u4e0b\u8868 1 \u6240\u793a\uff0c\u6b64\u66f8\u5c07\u76f8\u4f3c\u7684\u97f3\u6a02\u985e\u578b \u5206\u6210\u56db\u5927\u985e\u3002\u6b64\u8cc7\u6599\u96c6\u8f03\u7279\u5225\u7684\u9ede\u5728\u65bc\u63d0\u4f9b\u7684\u8072\u97f3\u6a94\u6848\u9577\u5ea6\u662f\u6574\u9996\u6b4c(Full Multitrack)\uff0c \u4e00\u822c\u4ee5\u5f80\u97f3\u6a02\u8cc7\u8a0a\u63a2\u52d8(Music Information Retrieval)\u7814\u7a76\u6240\u4f7f\u7528\u7684\u8cc7\u6599\u96c6\u5927\u591a\u662f 20~30 sec \u7684\u9577\u5ea6\uff0c\u9bae\u5c11\u6709\u63d0\u4f9b\u6574\u9996\u6b4c\u7684\u8cc7\u6599\u96c6\u3002\u6b64\u7279\u9ede\u6709\u5229\u65bc\u5e6b\u52a9\u6211\u5011\u5c0d\u65bc\u4e0d\u540c\u7684\u97f3\u6a02\u6a4b\u6bb5 (Music Section)\u53bb\u5efa\u7acb\u4e0d\u540c\u7684\u6a21\u578b\uff0c\u4f86\u8b93\u6700\u5f8c\u7684\u6a21\u578b\u80fd\u66f4\u9069\u7528\u65bc\u5be6\u969b\u7684\u60c5\u6cc1\uff0c\u6b64\u8cc7\u6599\u96c6 \u6240\u63d0\u4f9b \u7684\u6df7\u97f3 \u6210\u54c1 (Mixdown)\u4e00\u6a23\u4e5f\u662f \u6574\u9996 \u6b4c\u7684\u9577 \u5ea6\u3002 \u6a94 \u6848\u683c\u5f0f \u70ba\u7121\u640d WAV \u6a94 (uncompressed WAV files, 24bit and 44.1 kHz sample rate)\u3002 \u8868 1 \u97f3\u6a02\u985e\u578b\u7d71\u8a08\u8868 Genre \uff03 of Song \u7684\u5206\u8ecc\u6a94\u6848\u53ca\u6700\u5f8c\u5404\u6df7\u97f3\u7684\u53c3\u6578\u4f86\u9032\u884c\u76e3\u7763\u5f0f\u5b78\u7fd2(Supervised Learning)\u3002\u4f46\u662f\u5be6\u969b\u4e0a\u6df7 \u97f3\u53c3\u6578\u7684\u8cc7\u8a0a\u662f\u975e\u5e38\u96e3\u53d6\u5f97\u7684\u3002\u7531\u65bc\u6df7\u97f3\u5e2b\u4f7f\u7528\u7684\u8edf\u9ad4\u4e0d\u540c\uff0c\u4e0d\u540c\u7684\u5668\u6750/\u8edf\u9ad4\u6703\u6709\u4e0d \u540c\u7684\u8a2d\u5b9a\u3001\u57fa\u6e96\u53ca\u523b\u5ea6\uff0c\u800c\u4e14\u6df7\u97f3\u5e2b\u5728\u6df7\u97f3\u6642\u4e5f\u9bae\u5c11\u6703\u628a\u53c3\u6578\u8a18\u9304\u4e0b\u4f86\uff0c\u9019\u8b93\u5f97\u5230\u6df7\u97f3 \u53c3\u6578\u9019\u4e00\u985e\u7684\u8cc7\u8a0a\u7684\u53d6\u5f97\u8b8a\u5f97\u975e\u5e38\u56f0\u96e3\u3002\u70ba\u4e86\u4e4b\u5f8c\u7684\u76e3\u7763\u5f0f\u5b78\u7fd2\uff0c\u6211\u5011\u5fc5\u9808\u5148\u4f30\u8a08\u51fa\u8cc7 \u6599\u96c6\u4e2d\u6bcf\u9996\u6b4c\u7684\u6df7\u97f3\u53c3\u6578\uff0c\u5229 \u7528 \u539f \u59cb \u5206 \u8ecc\u6a94 \u6848 (Raw Multi-track) \u53ca\u6700\u5f8c\u6df7\u97f3\u6210\u54c1 (Mixdown)\u4f86\u4f30\u8a08\u51fa\u6bcf\u9996\u6b4c\u7684\u6df7\u97f3\u53c3\u6578\uff0c\u91dd\u5c0d\u6bcf\u4e00\u9996\u6b4c\u6c42\u5f97\u5176\u6df7\u97f3\u53c3\u6578\u7576\u4f5c\u4e4b\u5f8c\u76e3\u7763\u5f0f \u5b78\u7fd2(Supervised Learning)\u7684\u4f9d\u64da\u3002 \u70ba\u4e86\u5f9e\u539f\u59cb\u5206\u8ecc\u6a94\u6848\u4f30\u8a08\u51fa\u5176\u6df7\u97f3\u53c3\u6578\uff0c\u6211\u5011\u5047\u8a2d\u539f\u59cb\u5206\u8ecc\u8207\u6700\u5f8c\u6df7\u97f3\u7684\u6210\u54c1 (Final Mix)\u7684\u95dc\u4fc2\u662f\u4e00\u500b\u7dda\u6027\u7684\u7d44\u5408(Linear Combination)\uff0c\u5982\u4e0b(2)\u5f0f\u70ba\u4e00\u9996\u6b4c\u7684\u7dda\u6027\u7d44 \u5408\u95dc\u4fc2\u3002 \u03b1 1 U 1 +\u03b1 2 U 2 +\u2026..+\u03b1 k U k =V (2) \u03b1 i \u70ba\u7b2c i \u8ecc\u7684\u6df7\u97f3\u53c3\u6578\u6b0a\u91cd\uff0cU i = [u 1i , u 2i , \u2026 , u Ni ] T \u70ba\u7b2c i \u8ecc\u7279\u5fb5\u5411\u91cf\uff0c\uff36\u70ba\u6700\u5f8c\u6df7\u97f3\u7d50 \u679c\u7684\u7279\u5fb5\u5411\u91cf(Feature Vector)\uff0c\u6bcf\u4e00\u8ecc\u62bd\u53d6 N \u500b frames \u505a\u70ba\u5176\u4ee3\u8868\u3002\u5176\u4e2d\u5728\u4e0d\u540c\u7684\u6df7 \u983b\u7387(Frequency) \u983b\u7387\u53c3\u6578\u5373\u662f\u63a7\u5236\u6574\u9996\u6b4c\u4e2d\u5404\u97f3\u8ecc\u5728\u983b\u7387\u4e0a\u7684\u5e73\u8861\u3002\u5728\u6df7\u97f3\u904e\u7a0b\u4e2d\u70ba\u4e86\u4fee\u6b63\u97f3 \u8ecc\u7684\u97f3\u8272\u6216\u983b\u7387\u6642\u6703\u7528\u5230\u7b49\u5316\u5668\u4f86\u5e6b\u52a9\u6211\u5011\u5c0d\u8072\u97f3\u7684\u983b\u7387\u4f5c\u8abf\u6574\uff0c\u4e5f\u5c31\u662f\u8aaa\u983b\u7387\u53c3 \u6578\u5373\u662f\u7b49\u5316\u5668(Equalizer) \u53c3 \u6578 \u3002 \u5728 \u8a2d \u8a08 \u7b49 \u5316 \u5668 \u6642 \u6703 \u5148 \u5c07 \u6574 \u500b \u983b \u8b5c \u5207 \u6210 \u591a \u6bb5 (Multi-band)\uff0c\u5982\u5207\u6210\u4e09\u584a\u7684\u8a71\u5373\u662f\u9ad8\u983b\u3001\u4e2d\u983b\u3001\u4f4e\u983b\u3002\u63a5\u8457\u9078\u51fa\u5404\u983b\u6bb5\u7684\u4e2d\u5fc3\u983b\u7387 (Center Frequency)\u53ca\u983b\u5bec(bandwidth)\u5f8c\u5373\u5b8c\u6210\u8a2d\u8a08\u3002\u672c\u7bc7\u8ad6\u6587\u5728\u983b\u7387\u53c3\u6578\u65b9\u9762\u4e00\u6a23 \u63a1\u7528\u591a\u983b\u6bb5\u7b49\u5316\u65b9\u5f0f(Multi-band Equalization)\u4f86\u6a21\u64ec\u5be6\u969b\u7b49\u5316\u5668\u7684\u64cd\u4f5c\uff0c\u5373\u662f\u628a\u983b \u7387\u53c3\u6578\u4f30\u8a08\u7684\u554f\u984c\u5207\u6210\u4e86\u591a\u500b\u8072\u97f3\u53c3\u6578\u7684\u5b50\u554f\u984c\u3002\u5982\u4e0b\u5716 6 \u6240\u793a\u3002 \u5716 8 KDE \u793a\u610f\u5716 X Y* Y Y R k Y* R m . . . . . . . . . . . . . . . . ! (3)BACK to ORIGINAL S P ACE (1)P CA (2)Le a rning the Ma p \u5b89\u975c\u7684\u6a23\u672c(\u5373\u662f\u8aaa\u5728\u8a72\u6a23\u672c\u7684\u7576\u6642\u6709\u8f03\u591a\u8ecc\u97f3\u91cf\u662f\u8da8\u8fd1\u65bc 0)\u6642\u6703\u5c0e\u81f4\u5747\u65b9\u8aa4\u5dee\u503c\u98c6\u9ad8\uff0c \u5f62\u6210 Outlier\uff0c\u5982\u6416\u6efe\u985e\u4e2d\u7684\u7b2c 17 \u9996\u7684\u7d50\u679c\u3002 \u8868 3 \u526f\u6b4c\u97f3\u91cf\u6a21\u578b\u4ea4\u53c9\u9a57\u8b49\u7d50\u679c Rock/Metal POP Jazz/Country Alt Rock/Funk song1 0.137 0.043 0.145 0.027 song2 0.148 0.224 0.005 0.087 song3 0.192 0.057 0.018 0.040 song4 0.172 0.030 0.018 0.053 song5 0.205 0.056 0.138 0.038 song6 0.035 0.070 0.037 0.426 song7 0.166 0.067 0.017 0.018 song8 0.135 0.055 song9 0.052 song10 0.120 song11 0.199 song12 0.040 \u5716 10 \u97f3\u8ecc\u6bd4\u8f03\u5716 \u5716 11 \u4e0d\u540c m \u503c\u6bd4\u8f03 \u63a5\u8457\u6211\u5011\u5be6\u4f5c\u4e86\u76f8\u95dc\u7814\u7a76\u4e2d[3]\u6240\u4f7f\u7528\u7684\u591a\u7dda\u6027\u8ff4\u6b78(\u7121\u8003\u616e\u4f9d\u8cf4\u6027)\u8207\u672c\u7bc7\u8ad6\u6587\u7684 KDE(\u8003\u616e\u4f9d\u8cf4\u6027)\u7684\u65b9\u6cd5\u505a\u6bd4\u8f03\uff0c\u7d50\u679c\u5982\u4e0b\u5716 12\uff0c\u53ef\u767c\u73fe\u5728\u5927\u591a\u6578\u7684\u8ecc\u4e0a KDE \u6bd4\u591a\u7dda sub-band)\u5716 7 \u6df7\u97f3\u53c3\u6578\u9810\u6e2c\u793a\u610f\u5716 \u7528\u65bc\u672c\u7bc7\u8ad6\u6587\u7684\u4e3b\u984c\u4f86\u8aaa\uff0c\u6211\u5011\u7684\uff38\u5c31\u662f\u6211\u5011\u6bcf\u9996\u5206\u8ecc\u7684\u8072\u97f3\u7279\u5fb5\u5411\u91cf\uff0cY \u5c31\u662f\u6211 song13 0.092 song14 0.048 \u5728\u983b\u7387(Equalization)\u6a21\u578b\u65b9\u9762\u6211\u5011\u540c\u6a23\u4e5f\u4f5c\u4e86\u4ea4\u53c9\u9a57\u8b49\uff0c\u7531\u8868\u4e2d\u53ef\u77e5\u5927\u81f4\u4e0a\u5404\u5206\u8ecc \u6027\u8ff4\u6b78\u6709\u8f03\u597d\u7684\u8868\u73fe\u3002\u986f\u793a\u5176\u4f9d\u8cf4\u6027\u4f30\u8a08\u65b9\u5f0f\u8f03\u80fd\u8003\u91cf\u5404\u8ecc\u4e4b\u9593\u7684\u5e73\u8861\u3002 \u5716 13 \u985e\u5225\u4ea4\u53c9\u9a57\u8b49 \u7b2c\u4e8c\u6b65\u9a5f\u662f\u6838\u4f9d\u8cf4\u4f30\u8a08(Kernel Dependency Estimation)\u7684\u6a21\u578b\u5efa\u7acb\uff0c\u6211\u5011\u6703\u5229\u7528\u6bcf \u9996\u6b4c\u7684\u6df7\u97f3\u53c3\u6578\u6b0a\u91cd\u53ca\u7279\u5fb5\u5411\u91cf\u4f86\u7576\u4f5c\u8a13\u7df4\u6838\u4f9d\u8cf4\u4f30\u8a08(Kernel Dependency Estimation) \u7684\u4f9d\u64da\uff0c\u8a13\u7df4\u597d\u7684\u6a21\u578b\u5c07\u6703\u7528\u4f86\u9810\u6e2c\u5404\u500b\u6df7\u97f3\u53c3\u6578\u7684\u6b0a\u91cd\u3002\u6700\u5f8c\u4f9d\u7167\u6a21\u578b\u9810\u6e2c\u51fa\u7684\u6b0a\u91cd \u9032\u884c\u6df7\u97f3\u3002 \u5728\u63a5\u4e0b\u4f86\u7684\u7ae0\u7bc0\u6211\u5011\u6703\u8a73\u7d30\u4ecb\u7d39\u5404\u6b65\u9a5f\u7684\u505a\u6cd5\uff0c\u5728\u7ae0\u7bc0 3.1 \u6703\u5148\u5c0d\u6211\u5011\u6240\u4f7f\u7528\u7684\u8cc7 \u6599\u96c6(Data Set)\u505a\u4ecb\u7d39\u4ee5\u53ca\u524d\u8655\u7406\u7684\u90e8\u5206\u3002\u7ae0\u7bc0 3.2 \u6703\u4ecb\u7d39\u6211\u5011\u5982\u4f55\u5229\u7528\u6700\u5c0f\u5e73\u65b9\u6cd5\u4f86\u4f30 \u8a08\u5404\u500b\u6df7\u97f3\u53c3\u6578\u53ca\u70ba\u4f55\u9700\u8981\u505a\u6df7\u97f3\u53c3\u6578\u4f30\u8a08\u3002\u7ae0\u7bc0 3.3 \u6211\u5011\u6703\u4ecb\u7d39\u6838\u4f9d\u8cf4\u4f30\u8a08(\u4e0b\u9762\u7c21\u7a31 Alt Rock / Blues / Country Rock / Indie / Funk / Reggae 7 Rock / Punk / Metal 17 Pop / Singer-Songwriter 10 Acoustic / Jazz / Country / Orchestral 8 Total 42 \u97f3\u53c3\u6578\u6703\u7528\u4e0d\u540c\u7684\u8072\u97f3\u7279\u5fb5\uff0c\u4f8b\u5982\u5728\u97f3\u91cf\u53c3\u6578\u65b9\u9762\u6703\u63a1\u7528\u8072\u97f3\u7684\u65b9\u5747\u6839\u4f86\u7576\u4f5c\u8861\u91cf\u7684\u4f9d \u64da\uff0c\u983b\u7387\u53c3\u6578\u65b9\u9762\u6703\u63a1\u7528\u8072\u97f3\u7684\u983b\u8b5c(Spectrum)\u7b49\u7b49\u3002\u5229\u7528\u6b64\u7dda\u6027\u7d44\u5408\u7684\u95dc\u4fc2\uff0c\u6211\u5011\u53ef \u4ee5\u5229\u7528\u6700\u5c0f\u5e73\u65b9\u6cd5(Least Square Method)\u4f86\u4f30\u8a08\u51fa\u6df7\u97f3\u53c3\u6578 \u03b1 \u7684\u6578\u503c\uff0c\u6700\u5c0f\u5e73\u65b9\u6cd5(Least Square Method)\u662f\u4ee5\u89c0\u6e2c\u503c U \u8207\u9810\u6e2c\u503c \u00db \u4e4b\u5dee\u7684\u5e73\u65b9\u548c\u4f5c\u70ba\u6700\u4f73\u5316\u7684\u76ee\u6a19\u51fd\u6578 (RMS)\uff0c\u6bcf\u500b\u97f3\u6846\u9577\u5ea6\u7d04\u70ba 20 \u6beb\u79d2\uff0c\u5247(1)\u5373\u53ef\u8868\u793a\u70ba \u5716 6 \u591a\u983b\u6bb5\u983b\u8b5c \u5728\u9032\u884c\u983b\u7387\u53c3\u6578\u4f30\u8a08\u6642\uff0c\u6211\u5011\u6703\u5148\u7528\u5feb\u901f\u5085\u7acb\u8449\u8f49\u63db(Fast Fourier Transform)\u5148\u5f97\u5230 \u5404\u97f3\u8ecc\u7684\u983b\u8b5c(Spectrum)\uff0c\u63a5\u8457\u628a\u5404\u97f3\u8ecc\u7684\u983b\u8b5c\u4f9d\u7167\u6211\u5011\u9810\u5148\u5206\u6bb5\u7684\u983b\u7387\u5206\u5225\u53bb\u89e3 \u8ff4\u6b78\u554f\u984c(Regression Problem)\uff0c\u4ee5\u4f30\u8a08\u51fa\u5404\u983b\u6bb5\u5728\u5404\u8ecc\u9593\u7684\u5e73\u8861\u53c3\u6578\u3002\u5982\u4e0b(5)\u5f0f \u9019\u500b\u554f\u984c\u57fa\u672c\u4e0a\u662f\u53ef\u4ee5\u5229\u7528\u5e38\u898b\u7684\u53c3\u6578\u9810\u6e2c(Parameter Prediction)\u7684\u65b9\u6cd5\u5206\u5225\u53bb\u6c42 \u89e3\uff0c\u4f8b\u5982\u591a\u7dda\u6027\u8ff4\u6b78(Multiple Linear Regression)\u3001\u52d5\u614b\u7dda\u6027\u7cfb\u7d71(Linear Dynamic System); \u4f46\u662f\u7531\u65bc\u5728\u6df7\u97f3\u6642\u6bcf\u4e00\u8ecc\u9593\u7684\u53c3\u6578\u9593\u662f\u6709\u4e92\u76f8\u5f71\u97ff\u7684\uff0c\u5176\u4e2d\u4e00\u8ecc\u7684\u97f3\u91cf\u8b8a\u5927\u5f8c\u52e2\u5fc5\u5176\u4ed6 \u8ecc\u6703\u8b8a\u5c0f\u8072\u4e9b\uff0c\u5f7c\u6b64\u662f\u4f9d\u8cf4\u95dc\u4fc2\u7684(dependent)\u3002\u82e5\u7528\u5404\u5225\u8a13\u7df4\u6a21\u578b\u4f86\u9810\u6e2c\u6211\u5011\u7684\u6df7\u97f3\u53c3 \u6578\u7684\u8a71\uff0c\u6211\u5011\u5c07\u6703\u5f97\u5230 k \u500b\u7368\u7acb\u7684\u8ff4\u6b78\u6a21\u578b\uff0c\u9019 \u5c0d\u65bc\u6df7\u97f3\u7d50\u679c\u53ef\u80fd\u6703\u7522\u751f\u8f03\u4e0d\u548c\u8ae7\u7684\u5f71 \u97ff\u3002\u6240\u4ee5\u672c\u7bc7\u8ad6\u6587\u63a1\u7528\u4e86\u6838\u4f9d\u8cf4\u4f30\u8a08\u7684\u65b9\u5f0f\u4f86\u89e3\u6c7a\u4e0a\u8ff0\u7684\u554f\u984c\uff0c\u7528\u6838\u4f9d\u8cf4\u4f30\u8a08\u8a13\u7df4\u51fa\u6211 \u5011\u6700\u5f8c\u7684\u6a21\u578b\u3002\u63a5\u4e0b\u4f86\u5c07\u6703\u5148\u4ecb\u7d39\u6838\u4f9d\u8cf4\u4f30\u8a08\u7684\u57fa\u672c\u7cbe\u795e\u3002 song15 0.269 song16 0.748 song17 2.175 Mean 0.290 0.075 0.054 0.099 \u5716 9 \u662f\u5176\u4e2d\u4e00\u7a2e\u97f3\u6a02\u985e\u578b\u4e2d\u526f\u6b4c(Chorus)\u4ea4\u53c9\u9a57\u8b49\u7684\u8a73\u7d30\u5716\u8868\uff0cX \u8ef8\u4ee3\u8868\u5176\u96a8\u6a5f\u62bd \u7684\u6e96\u78ba\u7387\u7d04\u70ba 0.015\uff0c\u986f\u793a\u5728\u983b\u7387\u53c3\u6578\u9810\u6e2c\u4e0a\u6709\u8457\u8f03\u597d\u7684\u8868\u73fe\u3002 \u8868 4 \u983b\u7387\u6a21\u578b\u4ea4\u53c9\u9a57\u8b49 Rock/Metal POP Jazz/Country Alt Rock/Funk Kick 0.011 0.019 0.003 0.008 Snare 0.043 0.015 0.011 0.018 Hihat 0.003 0.001 0.020 0.021 \u4e94\u3001 \u7d50\u8ad6\u8207\u672a\u4f86\u5de5\u4f5c \u5728\u97f3\u6a02\u88fd\u4f5c\u4e0a\u6df7\u97f3\u662f\u975e\u5e38\u91cd\u8981\u7684\u4e00\u500b\u904e\u7a0b\uff0c\u97f3\u6a02\u6210\u54c1\u54c1\u8cea\u597d\u58de\uff0c\u53d6\u6c7a\u65bc\u6df7\u97f3\u662f\u5426\u6df7 \u5f97\u597d\u800c\u4e14\u6df7\u97f3\u727d\u6d89\u5230\u8a31\u591a\u97f3\u97ff\u53ca\u8072\u5b78\u5fc3\u7406\u5b78\u7684\u76f8\u95dc\u6280\u8853\u8207\u77e5\u8b58\uff0c\u975e\u5c08\u696d\u4eba\u58eb\u8981\u6df7\u51fa\u5c1a\u53ef \u5011\u539f\u5148\u4f30\u8a08\u51fa\u4f86\u7684\u6df7\u97f3\u53c3\u6578 Y\u2208R \u6295\u5f71\u56de\u539f\u672c\u7684 k \u7dad\u7a7a\u9593\u4e0a\u5373\u6c42\u51fa\u5404\u8ecc\u6df7\u97f3\u53c3\u6578\u6b0a\u91cd\u3002 \u6a23\u6e2c\u8a66\u7684\u8072\u97f3\u6a23\u672c(Sound Sample)\uff0c Y \u8ef8\u70ba\u5176\u6240\u5c0d\u61c9\u5747\u65b9\u8aa4\u5dee\uff0c\u5716\u8868\u4e2d\u540c\u6a23\u984f\u8272\u7684\u7dda\u689d TOM 0.008 0.017 0.013 0.007 DRUMROOM 0.013 0.004 0.036 \u7684\u6210\u54c1\u6709\u4e00\u5b9a\u7684\u96e3\u5ea6\uff0c\u6240\u4ee5\u672c\u7bc7\u8ad6\u6587\u63d0\u51fa\u4e00\u500b\u5229\u7528\u76e3\u7763\u5f0f\u5b78\u7fd2\u4f86\u9032\u884c\u81ea\u52d5\u591a\u8ecc\u6df7\u97f3\u7684\u7cfb 0.001 \u4ee3\u8868\u540c\u4e00\u9996\u6b4c\u7684\u8072\u97f3\u6a23\u672c\u3002\u53ef\u4ee5\u767c\u73fe KDE \u6240\u8a13\u7df4\u51fa\u7684\u6a21\u578b\u5728\u5c0d\u540c\u4e00\u9996\u6b4c\u7684\u8072\u97f3\u6a23\u672c\u6642 OVERHEAD 0.017 0.028 0.003 0.046 \u7d71\u3002\u8207\u5176\u4ed6\u7bc7\u76f8\u95dc\u7814\u7a76\u4e0d\u540c\u7684\u662f\u5176\u4e3b\u8981\u7684\u6838\u5fc3\u65b9\u6cd5\u662f\u6838\u4f9d\u8cf4\u4f30\u8a08(Kernel Dependency KDE)\u7684\u6838\u5fc3\u6982\u5ff5\u4ee5\u53ca\u5728\u672c\u7bc7\u8ad6\u6587\u7684\u554f\u984c\u4e2d\u8a72\u5982\u4f55\u8a2d\u8a08\u3002 \u5716 4 \u7cfb\u7d71\u67b6\u69cb\u5716 \u5716 5 \u6df7\u97f3\u53c3\u6578\u6a21\u578b Rock Intro. Ve rse Chorus Jazz Intro. Ve rse Chorus Pop Intro. Ve rse Chorus \u8cc7\u6599\u96c6\u524d\u8655\u7406\u7684\u90e8\u5206\uff0c\u7531\u65bc\u6b64\u8cc7\u6599\u96c6\u6db5\u84cb\u7684\u97f3\u6a02\u985e\u578b\u5305\u62ec Alt Rock/Blues \u3001 3.3.1 \u6838\u4f9d\u8cf4\u4f30\u8a08(Kernel Dependency Estimation) \u6703\u6709\u4e00\u81f4\u6548\u679c\u3002 PERCUSSION 0.008 0.006 0.025 0.005 BASS 0.011 0.007 0.004 Estimation)\uff0c\u5229\u7528\u6df7\u97f3\u53c3\u6578\u9593\u7684\u4f9d\u8cf4\u6027(dependency)\uff0c\u4f86\u505a\u6df7\u97f3\u53c3\u6578\u7684\u9810\u6e2c; \u53e6\u4e00\u500b\u8207\u5176 0.017 \u56db\u3001 \u5be6\u9a57 EG 0.006 0.019 0.024 0.005 \u4ed6\u76f8\u95dc\u8ad6\u6587\u4e0d\u540c\u7684\u662f\u8a13\u7df4\u7684\u55ae\u4f4d\u4e0d\u540c\uff0c\u7531\u65bc\u6df7\u97f3\u5176\u5be6\u662f\u975e\u5e38\u500b\u5225\u7684(Case By Case)\uff0c\u6df7 Rock/Punk\u3001Pop/Singer-Songwriter\u3001Acoustic/Jazz/Country \u7b49\uff0c\u6bcf\u500b\u97f3\u6a02\u985e\u578b\u6240\u4f7f\u7528\u7684\u914d \u5668\u6703\u7565\u6709\u4e0d\u540c\uff0c\u4f8b\u5982\u5728\u7235\u58eb\u6a02\u4e2d\u7ba1\u6a02\u5668\u7684\u4f7f\u7528\u6bd4\u4f8b\u6703\u6bd4\u6416\u6efe\u6a02\u4f86\u7684\u9ad8\u3002\u9019\u6a23\u6703\u9020\u6210\u4e4b\u5f8c AG 0.000 0.004 0.015 0.006 LEADVOX 0.021 0.013 0.005 \u5716 12 KDE \u8207 MLR \u6bd4\u8f03\u5716 \u97f3\u5e2b\u57fa\u672c\u4e0a\u90fd\u6703\u4f9d\u7167\u5404\u97f3\u8ecc\u7684\u7684\u8072\u97ff\u3001\u97f3\u6a02\u985e\u578b\u4e0d\u540c\u800c\u6709\u4e0d\u540c\u7684\u8655\u7406\u65b9\u5f0f; \u6240\u4ee5\u5728\u672c\u7bc7 0.025 (3) (5) \u6838\u4f9d\u8cf4\u4f30\u8a08(\u4ee5\u4e0b\u7c21\u7a31 KDE)\u662f\u4e00\u7a2e\u7528\u65bc\u5c0b\u627e\u8f38\u5165 X \u8207\u8f38\u51fa Y \u9593\u4f9d\u8cf4\u95dc\u4fc2\u7684\u5b78\u7fd2 \u5be6\u9a57\u5206\u70ba\u4e09\u500b\u90e8\u5206\u4f5c\u8a0e\u8ad6\uff0c\u7b2c\u4e00\u90e8\u5206\u7684\u5be6\u9a57\u662f\u8a55\u4f30\u7531\u6838\u4f9d\u8cf4\u4f30\u8a08\u6240\u5efa\u7acb\u7684\u6a21\u578b\u7684\u6b63 BACKVOX 0.002 0.003 0.032 0.017 \u8ad6\u6587\u7684\u6a21\u578b\u5efa\u7acb\u7684\u904e\u7a0b\u6211\u5011\u6703\u4f9d\u7167\u4e0d\u540c\u7684\u97f3\u6a02\u985e\u578b\u4ee5\u53ca\u6a4b\u6bb5\u5efa\u7acb\u4e0d\u540c\u7684\u6a21\u578b\uff0c\u4ee5\u65b9\u4fbf\u6700 \u7121\u6cd5\u5c07\u6b64\u8cc7\u6599\u96c6\u5957\u7528\u81f3\u6211\u5011\u7684\u6a21\u578b\u4e2d\u3002\u6240\u4ee5\u70ba\u4e86\u89e3\u6c7a\u6b64\u554f\u984c\u6211\u5011\u7d71\u8a08\u4e86\u5404\u97f3\u6a02\u985e\u578b\u6240\u4f7f \u7528\u7684\u914d\u5668\uff0c\u6c7a\u5b9a\u51fa\u6bcf\u500b\u985e\u578b\u57fa\u672c\u97f3\u8ecc(Basic Track)\u5982\u8868 2\uff0c\u6211\u5011\u5b9a\u7fa9\u4e86 12 \u7a2e\u97f3\u8ecc\u7684\u5f62\u614b\u3002 \u524d\u8655\u7406\u7684\u90e8\u5206\u6703\u5148\u5c0d\u5206\u8ecc\u505a\u97f3\u6a02\u683c\u5f0f\u4e0a\u7684\u8f49\u63db\uff0c\u5f85\u6bcf\u9996\u6b4c\u7684\u6df7\u97f3\u53c3\u6578\u6b0a\u91cd\u4f30\u8a08\u51fa\u4f86\u5f8c\u6703 \u5728\u6df7\u97f3\u6642\u7531\u65bc\u6df7\u97f3\u7684\u53c3\u6578\u773e\u591a\uff0c\u672c\u7bc7\u8ad6\u6587\u53ea\u8a0e\u8ad6\u4e86\u5176\u4e2d\u4e09\u500b\u6700\u91cd\u8981\u53c3\u6578\uff1aVolume\u3001 Frequency\u3001Pan\uff0c\u5728\u4f30\u8a08\u4e0d\u540c\u7684\u53c3\u6578\u6642\u6211\u5011\u6703\u4f7f\u7528\u4e0d\u540c\u7684\u8072\u97f3\u7279\u5fb5\u3002\u63a5\u4e0b\u4f86\u6703\u5c0d\u5404\u53c3\u6578 \u6240\u7528\u7684\u7279\u5fb5\u4f86\u505a\u4ecb\u7d39 \u6a02\u5668\u64fa\u4f4d(PANNING) \u6a02\u5668\u64fa\u4f4d\u5373\u662f\u6c7a\u5b9a\u8a72\u97f3\u8ecc\u5728\u5de6\u8072\u9053\u53ca\u53f3\u8072\u9053\u4e4b\u97f3\u91cf\u6bd4\u4f8b\uff0c\u5982\u5f0f(1)\u3002\u5728\u672c\u7bc7\u8ad6\u6587 \u67b6\u69cb\uff0cKDE \u7684\u6d41\u7a0b\u5982\u4e0b\u5716 8 \u6240\u793a\uff0c\u6b65\u9a5f\u4e3b\u8981\u5206\u70ba\u4e09\u500b\u6b65\u9a5f\uff1a 1. \u78ba\u6027; \u7b2c\u4e8c\u90e8\u5206\u662f\u8a55\u4f30 KDE \u4f9d\u8cf4\u6027(Dependency)\u7684\u6548\u679c\uff0c\u6bd4\u8f03\u4e0d\u540c\u7684 m \u503c\u5c0d\u6a21\u578b\u6b63\u78ba \u6027\u7684\u5f71\u97ff\u7a0b\u5ea6; \u7b2c\u4e09\u90e8\u5206\u662f\u8a55\u4f30\u4e0d\u540c\u7684\u97f3\u6a02\u985e\u578b\u6df7\u97f3\u65b9\u5f0f\u7684\u5dee\u7570\u6027\u3002\u5be6\u9a57\u8a55\u4f30\u7684\u65b9\u5f0f\u662f Mean 0.012 0.011 0.016 0.015 \u5f8c\u5be6\u969b\u4e0a\u7684\u5229\u7528\u3002\u7531\u5be6\u9a57\u7d50\u679c\u5f97\u53ef\u77e5\uff0c\u4e0d\u540c\u7684\u97f3\u6a02\u985e\u578b\u5176\u6df7\u97f3\u65b9\u5f0f\u662f\u6709\u5176\u7368\u7279\u6027\u7684\u3002 4.3 \u8de8\u985e\u578b\u6e2c\u8a66(Cross Genre Testing) \u672a\u4f86\u5de5\u4f5c\u65b9\u9762\uff0c\u7531\u65bc\u672c\u7bc7\u8ad6\u6587\u6709\u5206\u6210\u56db\u5927\u97f3\u6a02\u985e\u578b\u53bb\u505a\u6a21\u578b\u5efa\u7acb\uff0c\u5c0e\u81f4\u5404\u985e\u5225\u8a13\u7df4 1. \u5229\u7528\u4e00\u6b21\u6311\u4e00\u500b\u4ea4\u53c9\u9a57\u8b49\u5c0d\u540c\u4e00\u97f3\u6a02\u985e\u578b\u7684\u6b4c\u505a\u5747\u65b9\u8aa4\u5dee(Mean Square Error)\u8a55\u4f30\u3002\u5747 4.2 KDE \u65b9\u6cd5\u7684\u6548\u679c(Effect of KDE Method) \u5728\u7b2c\u4e00\u90e8\u5206\u7684\u5be6\u9a57\u6a21\u578b\u7684\u5efa\u7acb\u4ee5\u53ca\u6e2c\u8a66\u90fd\u4fb7\u9650\u5728\u540c\u4e00\u500b\u97f3\u6a02\u985e\u578b\u4e2d\uff0c\u5728\u7b2c\u4e09\u90e8\u5206\u7684 \u8cc7\u6599\u6709\u504f\u5c11\u7684\u50be\u5411\uff0c\u672a\u4f86\u5e0c\u671b\u5408\u4f75\u5176\u4ed6\u8cc7\u6599\u96c6\u4ee5\u88dc\u8db3\u5176\u6578\u91cf;\u518d\u8005\u7531\u65bc\u6df7\u97f3\u53c3\u6578\u662f\u975e\u5e38 \u505a\u57fa\u672c\u8ecc\u7684\u5408\u4f75(Track Merge)\uff0c\u5c07\u540c\u4e00\u985e\u578b\u7684\u97f3\u8ecc\u4f9d\u7167\u5176\u6b0a\u91cd\u5148\u5408\u4f75\u6210\u8a72\u985e\u578b\u57fa\u672c\u97f3\u8ecc\uff0c \u5c07\u540c\u4e00\u985e\u578b\u7684\u6a94\u6848\u5148\u5408\u4f75\u6210\u4e00\u500b;\u4f8b\u5982\u5728\u67d0\u4e00\u9996\u6b4c\u4e2d\u5409\u4ed6\u9304\u4e86\u5169\u628a\uff0c\u6211\u5011\u6703\u5148\u5c07\u9019\u5169\u628a \u7684\u9304\u97f3\u5408\u4f75\u6210\u4e00\u500b\uff0c\u4ee5\u65b9\u4fbf\u4e4b\u5f8c\u8a13\u7df4\u53ca\u6e2c\u8a66\u8a72\u985e\u578b\u7684\u97f3\u6a02\u3002 \u8868 2 \u57fa\u672c\u8ecc\u8868 12 Basic Track Type (1)Kick (2)Snare (3)Hihat (4)TOM (5)DrumRoom (6)OVERHEAD (7)PERCUSSION (8)BASS (9)Electric Guitar (10)Acoustic Guitar (11)LeadVox. (12)BackVox \u6211\u5011\u5c07\u64fa\u4f4d\u53c3\u6578\u7684\u554f\u984c\u8f49\u5316\u6210\u524d\u4e00\u7ae0\u7bc0\u97f3\u91cf\u53c3\u6578\u4f30\u8a08\u7684\u554f\u984c\uff0c\u5373\u5de6\u8072\u9053\u505a\u4e00\u6b21\u97f3\u91cf \u65b9\u8aa4\u5dee\u7684\u8a08\u7b97\u65b9\u5f0f\u5982\u4e0b\uff1a \u5be6\u9a57\uff0c\u6211\u5011\u60f3\u8981\u8a55\u4f30\u4e0d\u540c\u7684\u97f3\u6a02\u98a8\u683c\u7684\u97f3\u6a02\u662f\u5426\u6709\u5176\u6df7\u97f3\u7279\u8272\uff0c\u6211\u5011\u5c07\u6703\u5957\u7528\u7531\u4e0d\u540c\u985e \u96e3\u53d6\u5f97\u7684\uff0c\u672c\u7bc7\u8ad6\u6587\u662f\u63a1\u7528\u4f30\u8a08\u7684\u65b9\u5f0f\u4f30\u8a08\u5176\u8cc7\u6599\u96c6\u4e2d\u7684\u6df7\u97f3\u53c3\u6578\u6b0a\u91cd\uff0c\u672a\u4f86\u53ef\u4ee5\u6539\u7528 \u97f3\u91cf (VOLUME) \u97f3\u91cf\u53c3\u6578\u4e5f\u88ab\u7a31\u4f5c\u589e\u76ca(Gain)\uff0c\u4e3b\u8981\u5728\u63a7\u5236\u6bcf\u500b\u97f3\u8ecc\u9593\u7684\u97f3\u91cf\u4f7f\u5f97\u6574\u9ad4\u97f3\u91cf\u9054 \u6210\u5e73\u8861\u3002\u672c\u7bc7\u8ad6\u6587\u4f7f\u7528\u4e86\u65b9\u5747\u6839(Root Mean Square)\u7684\u65b9\u5f0f\u53bb\u6e2c\u91cf\u540c\u4e00\u500b\u97f3\u6846(Frame) \u4e2d\u5404\u97f3\u8ecc\u7684\u8072\u58d3(Sound Pressure Level)\uff0c\u4ee5\u6b64\u4f5c\u70ba\u97f3\u91cf\u53c3\u6578\u7684\u7279\u5fb5\u5411\u91cf\uff0c\u5beb\u6210\u77e9\u9663 \u5f62\u5f0f\u5982(4)\u5f0f\u3002G k \u70ba\u7b2c k \u8ecc\u7684\u6b0a\u91cd\uff0cN \u70ba\u97f3\u6846\u7684\u9577\u5ea6\u3002 \u53c3\u6578\u4f30\u8a08\uff0c\u53f3\u8072\u9053\u505a\u4e00\u6b21\u97f3\u91cf\u53c3\u6578\u4f30\u8a08\uff0c\u9019\u6a23\u5373\u53ef\u6c7a\u5b9a\u8a72\u97f3\u8ecc\u5728\u5de6\u53f3\u8072\u9053\u4e4b\u6bd4\u4f8b\uff0c \u5982\u4e0b\u5f0f(6)\u3001(7)\u6240\u793a\u3002 \u03b1 1 u L1 +\u03b1 2 u L2 +\u2026..+\u03b1 k u Lk =V (6) \u03b1 1 u R1 +\u03b1 2 u R2 +\u2026..+\u03b1 k u RK =V (7) 3.3 \u6838\u4f9d\u8cf4\u4f30\u8a08\u6a21\u578b\u5efa\u7acb(Kernel Dependency Estimation Model) \u4f30\u8a08\u5b8c\u8cc7\u6599\u96c6\u4e2d\u6bcf\u9996\u6b4c\u7684\u4e09\u7a2e\u6df7\u97f3\u53c3\u6578(Volume\u3001EQ\u3001Panning)\u5f8c\uff0c\u6211\u5011\u6709\u4e86\u6bcf\u9996 \u6b4c\u7684\u7279\u5fb5\u5411\u91cf(X)\uff0c\u53ca\u6bcf\u9996\u6b4c\u7684\u6df7\u97f3\u53c3\u6578(Y)\uff0c\u5373\u53ef\u85c9\u7531\u5b78\u7fd2\u51fa X, Y \u9593\u7684\u95dc\u4fc2\u5efa\u7acb\u6211\u5011 \u5c0d\u8f38\u51fa Y \u505a\u964d\u7dad\u7684\u52d5\u4f5c\uff0c\u6295\u5f71\u81f3\u8f03\u4f4e\u7684\u7dad\u5ea6\u4e0b\u5f62\u6210 Y'\u3002 2. \u5b78\u7fd2(Learning The Map)\uff1a\u5c0d\u6bcf\u500b\u57fa\u790e\u539f\u4ef6(Principal Component)j\uff0c1\u2264j\u2264m \uff0c\u6211\u5011\u6703 \u5b78\u7fd2\u4e00\u500b\u5c0d\u61c9\u51fd\u6578(Mapping Function)r j (X)\uff0c\u5c0d\u61c9\u51fd\u6578 r \u6703\u5c07\u6703\u628a X \u5c0d\u61c9\u81f3 Y'\u7b2c j \u500b \u57fa\u790e\u539f\u4ef6\uff0c\u5373\u662f\u5728\u8f03\u4f4e\u7684\u7a7a\u9593\u4e0a\u53bb\u89e3 m \u500b\u8ff4\u6b78\u554f\u984c\u3002 3. \u9810\u6e2c(Prediction)\uff1a\u5c0d\u65bc\u4e00\u500b\u65b0\u7684\u8f38\u5165 x'\uff0c\u6211\u5011\u53ef\u4ee5\u5229\u7528\u5148\u524d\u5b78\u7fd2\u597d\u7684\u5c0d\u61c9\u51fd\u6578\u6c42\u51fa y'' (Mean Square Error)\uff0c\u6e2c\u8a66\u6642\u6703\u5148\u5c0d\u6e2c\u8a66\u6b4c\u66f2\u505a\u96a8\u6a5f\u62bd\u6a23 3000 \u500b\u6a23\u672c\u4f86\u505a\u6e2c\u8a66\uff0cm \u503c\u7686 \u97f3(Multi-track Recording)\u6642\uff0c\u55ae\u4e00\u8ecc\u4e5f\u6703\u53c3\u96dc\u8457\u5176\u4ed6\u8ecc\u7684\u8072\u97f3(\u4e32\u97f3)\uff0c\u5982 Overhead\u3001 \u97f3\u65b9\u5f0f\uff0c\u6240\u8a13\u7df4\u51fa\u4f86\u7684\u6a21\u578b\u6709\u5176\u7368\u7279\u6027\u3002 = [ r 1 (x') r 2 (x')\u2026..r m (x')]\uff0cy'' \u2208R m \u3002\u6700\u5f8c\u518d\u5c07 Y''\u6295\u5f71\u56de\u539f\u672c\u7684\u7a7a\u9593\u4e0a\uff0c\u5373\u6c42\u51fa X'\u6240 = 1 \u2211 (\u03b1 \u2212 \u03b1 ) 2 =1 \u578b\u6240\u8a13\u7df4\u51fa\u7684\u6a21\u578b\u4f86\u770b\u770b\u5176\u6548\u679c\u662f\u5426\u6709\u5dee\u5225\u3002\u7d50\u679c\u5982\u4e0b\u5716 13\uff0cX \u8ef8\u5206\u5225\u662f\u5148\u524d\u5b9a\u7fa9\u7684\u57fa \u76f8\u95dc\u7814\u7a76\u7684\u4f30\u8a08\u65b9\u5f0f\u6216\u662f\u5be6\u969b\u6536\u96c6\u6df7\u97f3\u53c3\u6578\u4ee5\u8b93\u6700\u5f8c\u7684\u6df7\u97f3\u6a21\u578b\u80fd\u66f4\u8cbc\u8fd1\u5be6\u969b\u4e0a\u7684\u60c5 (8) \u03b1\u4ee3\u8868\u7531 KDE \u6a21\u578b\u6240\u9810\u6e2c\u7684\u6df7\u97f3\u53c3\u6578\u6b0a\u91cd\uff0c\u03b1 \u662f\u7531\u53c3\u6578\u4f30\u8a08\u5f97\u51fa\u5be6\u969b\u7684\u6df7\u97f3\u53c3\u6578\u6b0a\u91cd\u503c\u3002 4.1 \u3127\u6b21\u4e00\u500b\u4ea4\u53c9\u6e2c\u8a66(Leave-one-out Cross Validation) \u8868 3 \u70ba\u6bcf\u500b\u97f3\u6a02\u985e\u578b\u526f\u6b4c\u97f3\u91cf\u53c3\u6578\u7684\u4e00\u6b21\u4e00\u500b\u4ea4\u53c9\u9a57\u8b49 (Leave-one-out Cross Validation)\u7684\u7d50\u679c\uff0c\u5176\u503c\u70ba KDE \u6a21\u578b\u6240\u9810\u6e2c\u7684\u503c\u8207\u5be6\u969b\u6df7\u97f3\u53c3\u6578\u6b0a\u91cd\u7684\u5747\u65b9\u8aa4\u5dee\u503c \u5716 9 POP/Sing-Song writer \u5728\u53e6\u4e00\u65b9\u9762\u7531\u97f3\u8ecc\u7684\u89c0\u9ede\u4f86\u770b\uff0c\u5982\u5716 10\uff0c\u6211\u5011\u53ef\u4ee5\u767c\u73fe\u5728 Kick\u3001DrumRoom\u3001 Overhead \u7b49\u97f3\u8ecc\u4e0a\u5404\u985e\u578b\u6703\u6709\u8f03\u5927\u7684\u8aa4\u5dee\u51fa\u73fe\uff0c\u5176\u539f\u56e0\u8981\u6b78\u548e\u65bc\u5728\u5be6\u969b\u4e0a\u591a\u8ecc\u540c\u6b65\u9304 \u5c0d\u65bc\u6df7\u97f3\u53c3\u6578\u9810\u6e2c\u7684\u5e6b\u52a9\u3002 \u6cc1\uff0c\u4f8b\u5982\u5efa\u69cb\u4e00\u500b\u7dda\u4e0a\u6df7\u97f3\u7cfb\u7d71\u7684\u65b9\u5f0f\u8b93\u4f7f\u7528\u8005\u5be6\u969b\u6df7\u97f3\u8a18\u9304\u5176\u6df7\u97f3\u53c3\u6578\u3002\u53e6\u5916\u5728 KDE \u65b9\u6cd5\u7684\u90e8\u5206\u7531\u65bc\u672c\u7bc7\u672a\u5957\u7528\u5176\u6838\u51fd\u6578(Kernel function)\u7684\u90e8\u5206\uff0c\u672a\u4f86\u53ef\u4ee5\u5728\u505a\u6295\u5f71\u524d\u5148\u5957 \u7528\u6838\u51fd\u6578\u4f86\u63d0\u5347\u6df7\u97f3\u6a21\u578b\u7684\u6548\u679c\u3002\u5be6\u9a57\u90e8\u5206\u8a55\u4f30\u7684\u5c0d\u8c61\u4e5f\u53ef\u4ee5\u65b0\u589e\u8ddf\u8cc7\u6599\u96c6\u7684\u6df7\u97f3\u6210\u54c1 \u505a\u6bd4\u8f03\u8a66\u8457\u770b\u770b\u76ee\u524d\u7684\u6df7\u97f3\u6a21\u578b\u8207\u5be6\u969b\u4e0a\u7684\u5dee\u8ddd\u5982\u4f55\u3002 \u672c\u8ecc\uff0cY \u8ef8\u662f\u5176\u5c0d\u61c9\u7684\u5747\u65b9\u8aa4\u5dee\u503c\uff0c\u7576\u4e2d\u6e2c\u8a66\u8cc7\u6599\u70ba \u986f\u793a\u4e0d\u592a\u9069\u5408\u3002\u6211\u5011\u4e5f\u767c\u73fe\u97f3\u6a02\u985e\u578b\u76f8\u4f3c\u7684\u6b4c\u5176\u6df7\u97f3\u65b9\u5f0f\u6703\u8f03\u76f8\u8fd1\u5982\u5716\u4e2d ROCK \u8207 Alt ROCK \u7684\u5747\u65b9\u8aa4\u5dee\u503c\u8f03\u70ba\u63a5\u8fd1\u3002\u7d93\u7531\u6b64\u5be6\u9a57\u6211\u5011\u53ef\u4ee5\u5f97\u77e5\u4e0d\u540c\u7684\u97f3\u6a02\u985e\u578b\u6709\u5176\u4e0d\u540c\u7684\u6df7 \u53c3\u8003\u6587\u737b
\u800c\u975e\u5168\u90e8\u7684\u6a02\u5668\u90fd\u64e0\u5728\u4e00\u8d77\u3002\u5728\u5be6\u4f5c\u4e0a\u5373\u662f\u5206\u5225\u8abf\u6574\u5de6\u8072\u9053\u8207\u53f3\u8072\u9053\u7684\u97f3\u91cf\uff0c\u5982\u4e0b(1) \u5f0f \u97f3\u3002 \u5efa\u7acb\u4e00\u500b\u6a21\u578b\uff0c\u85c9\u7531\u6bcf\u4e00\u8ecc\u7684\u7279\u5fb5\u4f86\u9810\u6e2c\u5176\u53c3\u6578\u7684\u503c\u3002\u6240\u4ee5\u8a13\u7df4\u7684\u904e\u7a0b\u4e2d\u52e2\u5fc5\u9700\u8981\u539f\u59cb \u6700\u5f8c\u7684\u6df7\u97f3\u6a21\u578b\u3002\u5728\u6a21\u578b\u5efa\u7acb\u7684\u90e8\u5206\u6211\u5011\u6703\u4f9d\u7167\u4e0d\u540c\u7684\u97f3\u6a02\u985e\u578b\uff0c\u4e0d\u540c\u7684\u97f3\u6a02\u6a4b\u6bb5\u5efa\u7acb \u5c0d\u61c9\u7684\uff39\u503c\u3002 \u70ba 8\u3002\u7531\u8868\u53ef\u77e5\u7531 KDE \u6240\u8a13\u7df4\u51fa\u7684\u6a21\u578b\u5747\u65b9\u8aa4\u5dee\u5e73\u5747\u5927\u7d04\u843d\u5728 0.129 \u5de6\u53f3\u7684\u6578\u503c\uff0c\u9810\u6e2c DrumRoom \u7b49\u8ecc\u6703\u5305\u542b kick\u3001snare \u7b49\u5176\u4ed6\u9f13\u7d44\u7684\u6a02\u5668\u3002\u6b64\u539f\u56e0\u9055\u53cd\u4e86\u7576\u521d\u7814\u7a76\u65b9\u6cd5\u4e00 (4) \u4e00\u7d44\u6a21\u578b\uff0c\u5176\u4e2d\u4e00\u7d44\u6a21\u578b\u5305\u62ec\u4e86\u97f3\u91cf\u6a21\u578b\u5169\u500b(\u5de6\u8072\u9053\u3001\u53f3\u8072\u9053)\uff0c\u983b\u7387\u6a21\u578b\u4e94\u500b(5 \u51fa\u7684\u6b0a\u91cd\u6709\u8457\u9084\u4e0d\u932f\u7684\u6e96\u78ba\u5ea6\uff0c\u4f46\u7531\u65bc\u662f\u63a1\u7528\u96a8\u6a5f\u62bd\u6a23\u7684\u65b9\u5f0f\uff0c\u82e5\u62bd\u5230\u7684\u8072\u97f3\u6a23\u672c\u70ba\u8f03 \u958b\u59cb\u7684\u5047\u8a2d: \u6211\u5011\u5047\u8a2d\u5206\u8ecc\u5373\u6df7\u97f3\u6210\u54c1\u9593\u662f\u500b\u7dda\u6027\u7d44\u5408\u7684\u95dc\u4fc2\uff0c\u5206\u8ecc\u9593\u5fc5\u9808\u8981\u662f\u7dda\u6027\u7368
", "text": "\u3002\u5229\u7528\u6b64\u6a21\u578b\u5efa\u7acb\u7684\u65b9\u6cd5\u8b93\u6700\u5f8c\u7684\u6210\u54c1\u66f4\u80fd\u61c9\u7528\u5728\u5be6\u969b\u7684\u6b4c\u66f2\u4e2d\uff0c\u8a13\u7df4\u6a21\u578b\u7684 \u6a23\u672c\u6578\u6211\u5011\u81ea\u5404\u8a13\u7df4\u6b4c\u66f2\u4e2d\u96a8\u6a5f\u62bd\u53d6 3000 \u500b\u6a23\u672c\u4f86\u7576\u4f5c\u6211\u5011\u7684\u8a13\u7df4\u8cc7\u6599\u3002\u5982\u5716 7\uff0cu nk \u70ba\u8a72\u6a21\u578b\u7684\u7b2c n \u500b\u6a23\u672c\u7684\u7b2c k \u8ecc\u7684\u8072\u97f3\u7279\u5fb5\u503c\uff0c\u03b1 nk \u70ba\u7b2c n \u500b\u6a23\u672c\u4e4b\u7b2c\uff2b\u8ecc\u7684\u6df7\u97f3\u53c3\u6578 \u6b0a\u91cd\u503c;K \u70ba\u5206\u8ecc\u7684\u500b\u6578\uff0c\u5728\u672c\u7bc7\u8ad6\u6587\u70ba 12\u3002 \u7b2c\u4e8c\u90e8\u5206\u7684\u5be6\u9a57\u662f\u8981\u4f86\u8a55\u4f30 KDE \u5373\u5176\u4f9d\u8cf4\u6027\u7684\u6210\u6548\uff0c\u9996\u5148\u6211\u5011\u8a0e\u8ad6\u4e86\u4e0d\u540c\u7684 m \u503c \u6240\u5e36\u4f86\u7684\u5f71\u97ff\uff0c\u7d50\u679c\u5982\u4e0b\u5716 11\uff0c\u70ba ROCK \u5728\u526f\u6b4c\u6642\u4e0d\u540c m \u503c\u7684\u5747\u65b9\u5dee\uff0cX \u8ef8\u70ba\u4e0d\u540c\u7684 m \u503c\uff0cY \u70ba\u6240\u5c0d\u61c9\u5176\u6df7\u97f3\u3002\u6211\u5011\u53ef\u4ee5\u767c\u73fe\u7576 m \u503c\u8d8a\u4f4e KDE \u6a21\u578b\u6703\u6709\u8f03\u597d\u7684\u7d50\u679c\uff0c\u8f03\u4f4e \u7684 m \u503c\u4e5f\u53ef\u52a0\u901f KDE \u7684\u8a08\u7b97(eg.PCA space \u5f9e R 12 \u964d\u4f4e\u70ba R 8 )\u3002\u9019\u7d50\u679c\u4e5f\u986f\u793a\u4e86\u4f9d\u8cf4\u6027 ROCK \u9019\u4e00\u985e\u7684\u6b4c\uff0c\u4e00\u5171\u6709 17 \u9996\uff0c \u5716\u4e2d\u7684\u7dda\u689d\u70ba\u5206\u5225\u7528 POP\u3001jazz\u3001Alt Rock \u6240\u8a13\u7df4\u51fa\u7684\u6a21\u578b\u5957\u7528\u81f3 ROCK \u985e\u5225\u7684\u6e2c\u8a66\u7d50 \u679c\u3002\u6211\u5011\u53ef\u4ee5\u767c\u73fe\u5c07\u5225\u7684\u985e\u5225\u7684\u6a21\u578b\u5957\u7528\u5728\u4e0d\u540c\u985e\u578b\u7684\u6b4c\u6642\u6703\u5c0e\u81f4\u6a21\u578b\u7684\u6b63\u78ba\u5ea6\u4e0b\u964d\u3001 \u8aa4\u5dee\u589e\u5927\uff0c\u5982\u5716\u4e2d\u7684 POP \u8207 JAZZ \u7684\u7d50\u679c\uff0c\u5169\u985e\u5225\u7684\u6a21\u578b\u5957\u7528\u5728 ROCK \u97f3\u6a02\u4e0a\u5176\u7d50\u679c" } } } }