{ "paper_id": "O15-1027", "header": { "generated_with": "S2ORC 1.0.0", "date_generated": "2023-01-19T08:09:59.865529Z" }, "title": "Speech Emotion Recognition via Nonlinear Dynamical Features", "authors": [ { "first": "Chu-Hsuan", "middle": [], "last": "\u6797\u7af9\u8431", "suffix": "", "affiliation": { "laboratory": "", "institution": "\u7f8e\u5f8b\u5be6\u696d\u80a1\u4efd\u6709\u9650\u516c\u53f8 Merry Electronics Co", "location": { "settlement": "Ltd" } }, "email": "" }, { "first": "", "middle": [], "last": "Lin", "suffix": "", "affiliation": { "laboratory": "", "institution": "\u7f8e\u5f8b\u5be6\u696d\u80a1\u4efd\u6709\u9650\u516c\u53f8 Merry Electronics Co", "location": { "settlement": "Ltd" } }, "email": "" }, { "first": "Yen-Sheng", "middle": [], "last": "\u9673\u708e\u751f", "suffix": "", "affiliation": {}, "email": "" }, { "first": "", "middle": [], "last": "Chen", "suffix": "", "affiliation": {}, "email": "" } ], "year": "", "venue": null, "identifiers": {}, "abstract": "This study is focus on speech emotion recognition through machine learning method. We add two nonlinear dynamical features: Shannon entropy and curvature index, of each frame other than the traditional features such as pitch, formant, energy, MFCCs. After feature extraction, Fisher discriminant ratio and Genetic algorithm were applied in order to reduce the number of features. We use SVM classifier and cross validation method to discriminate seven emotions in Berlin emotion database. The analyzed results after adding of the nonlinear features show that the emotion recognition rates were 88.89% and 86.21% for male and female, respectively.", "pdf_parse": { "paper_id": "O15-1027", "_pdf_hash": "", "abstract": [ { "text": "This study is focus on speech emotion recognition through machine learning method. We add two nonlinear dynamical features: Shannon entropy and curvature index, of each frame other than the traditional features such as pitch, formant, energy, MFCCs. After feature extraction, Fisher discriminant ratio and Genetic algorithm were applied in order to reduce the number of features. We use SVM classifier and cross validation method to discriminate seven emotions in Berlin emotion database. The analyzed results after adding of the nonlinear features show that the emotion recognition rates were 88.89% and 86.21% for male and female, respectively.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Abstract", "sec_num": null } ], "body_text": [], "back_matter": [], "bib_entries": { "BIBREF0": { "ref_id": "b0", "title": "\u8bed\u97f3\u60c5\u611f\u8bc6\u522b\u7814\u7a76\u8fdb\u5c55\u7efc\u8ff0", "authors": [ { "first": "", "middle": [], "last": "\u97e9\u6587\u9759", "suffix": "" } ], "year": 2014, "venue": "\u8f6f\u4ef6\u5b66\u62a5", "volume": "25", "issue": "", "pages": "37--50", "other_ids": {}, "num": null, "urls": [], "raw_text": "\u97e9\u6587\u9759, et al. \"\u8bed\u97f3\u60c5\u611f\u8bc6\u522b\u7814\u7a76\u8fdb\u5c55\u7efc\u8ff0.\" \u8f6f\u4ef6\u5b66\u62a5 25.1 (2014): 37-50.", "links": null }, "BIBREF1": { "ref_id": "b1", "title": "Research on key issues of Mandarin speech emotion recognition", "authors": [ { "first": "B", "middle": [], "last": "Xie", "suffix": "" } ], "year": 2006, "venue": "", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "Xie B. Research on key issues of Mandarin speech emotion recognition [Ph.D. Thesis]. Hangzhou: Zhejiang University, 2006 (in Chinese with English abstract).", "links": null }, "BIBREF2": { "ref_id": "b2", "title": "Emotion recognition in human-computer interaction", "authors": [ { "first": "R", "middle": [], "last": "Cowie", "suffix": "" }, { "first": "E", "middle": [], "last": "Douglas-Cowie", "suffix": "" }, { "first": "N", "middle": [], "last": "Tsapatsoulis", "suffix": "" }, { "first": "G", "middle": [], "last": "Votsis", "suffix": "" }, { "first": "S", "middle": [], "last": "Kollias", "suffix": "" }, { "first": "W", "middle": [], "last": "Fellenz", "suffix": "" }, { "first": "J", "middle": [ "G" ], "last": "Taylor", "suffix": "" } ], "year": null, "venue": "Proc. of the IEEE Signal Processing Magazine. 2001. 32\u221280", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "Cowie R, Douglas-Cowie E, Tsapatsoulis N, Votsis G, Kollias S, Fellenz W, Taylor JG. Emotion recognition in human-computer interaction. In: Proc. of the IEEE Signal Processing Magazine. 2001. 32\u221280. http://www.signalprocessingsociety.org/", "links": null }, "BIBREF3": { "ref_id": "b3", "title": "Survey on speech emotion recognition: Features, classification schemes, and databases", "authors": [ { "first": "Mohamed", "middle": [ "S" ], "last": "Moataz El Ayadi", "suffix": "" }, { "first": "Fakhri", "middle": [], "last": "Kamel", "suffix": "" }, { "first": "", "middle": [], "last": "Karray", "suffix": "" } ], "year": 2011, "venue": "Pattern Recognition", "volume": "44", "issue": "", "pages": "572--587", "other_ids": {}, "num": null, "urls": [], "raw_text": "Moataz El Ayadi, Mohamed S. Kamel, and Fakhri Karray. \"Survey on speech emotion recognition: Features, classification schemes, and databases.\"Pattern Recognition 44.3 (2011): 572-587.", "links": null }, "BIBREF4": { "ref_id": "b4", "title": "Automatic speech emotion recognition using modulation spectral features", "authors": [ { "first": "Siqing", "middle": [], "last": "Wu", "suffix": "" }, { "first": "H", "middle": [], "last": "Tiago", "suffix": "" }, { "first": "Wai-Yip", "middle": [], "last": "Falk", "suffix": "" }, { "first": "", "middle": [], "last": "Chan", "suffix": "" } ], "year": 2011, "venue": "Speech communication", "volume": "53", "issue": "", "pages": "768--785", "other_ids": {}, "num": null, "urls": [], "raw_text": "Siqing Wu, Tiago H. Falk, and Wai-Yip Chan. \"Automatic speech emotion recognition using modulation spectral features.\" Speech communication 53.5 (2011): 768-785.", "links": null }, "BIBREF5": { "ref_id": "b5", "title": "Nonlinear dynamics characterization of emotional speech", "authors": [ { "first": "Patricia", "middle": [], "last": "Henr\u00edquez", "suffix": "" } ], "year": 2014, "venue": "Neurocomputing", "volume": "132", "issue": "", "pages": "126--135", "other_ids": {}, "num": null, "urls": [], "raw_text": "Patricia Henr\u00edquez, et al. \"Nonlinear dynamics characterization of emotional speech.\" Neurocomputing 132 (2014): 126-135.", "links": null }, "BIBREF6": { "ref_id": "b6", "title": "Speech emotion recognition using non-linear dynamics features", "authors": [ { "first": "Ali", "middle": [], "last": "Shahzadi", "suffix": "" } ], "year": 2013, "venue": "Turkish Journal of Electrical Engineering & Computer Sciences", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "Ali Shahzadi, et al. \"Speech emotion recognition using non-linear dynamics features.\" Turkish Journal of Electrical Engineering & Computer Sciences. doi10 (2013).", "links": null }, "BIBREF7": { "ref_id": "b7", "title": "A database of German emotional speech", "authors": [ { "first": "Felix", "middle": [], "last": "Burkhardt", "suffix": "" } ], "year": 2005, "venue": "Interspeech", "volume": "5", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "Burkhardt, Felix, et al. \"A database of German emotional speech.\" Interspeech. Vol. 5. 2005.", "links": null }, "BIBREF8": { "ref_id": "b8", "title": "The Curvature Index and Synchronization of Dynamical Systems", "authors": [ { "first": "Yen-Sheng", "middle": [], "last": "Chen", "suffix": "" }, { "first": "Chien-Cheng", "middle": [], "last": "Chang", "suffix": "" } ], "year": 2012, "venue": "CHAOS", "volume": "22", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "Yen-Sheng Chen and Chien-Cheng Chang, 2012, \"The Curvature Index and Synchronization of Dynamical Systems\", CHAOS 22, 023131.", "links": null }, "BIBREF9": { "ref_id": "b9", "title": "A feature selection method based on fisher's discriminant ratio for text sentiment classification", "authors": [ { "first": "Suge", "middle": [], "last": "Wang", "suffix": "" } ], "year": 2009, "venue": "", "volume": "", "issue": "", "pages": "88--97", "other_ids": {}, "num": null, "urls": [], "raw_text": "Suge Wang, et al. \"A feature selection method based on fisher's discriminant ratio for text sentiment classification.\" Web Information Systems and Mining. Springer Berlin Heidelberg, 2009. 88-97.", "links": null }, "BIBREF10": { "ref_id": "b10", "title": "An introduction to genetic algorithms", "authors": [ { "first": "Melanie", "middle": [], "last": "Mitchell", "suffix": "" } ], "year": 1996, "venue": "", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "Melanie Mitchell. An introduction to genetic algorithms. MIT press, 1996.", "links": null }, "BIBREF11": { "ref_id": "b11", "title": "A GA-based feature selection and parameters optimizationfor support vector machines", "authors": [ { "first": "Cheng-Lung", "middle": [], "last": "Huang", "suffix": "" }, { "first": "Chieh-Jen", "middle": [], "last": "Wang", "suffix": "" } ], "year": 2006, "venue": "Expert Systems with applications", "volume": "31", "issue": "2", "pages": "231--240", "other_ids": {}, "num": null, "urls": [], "raw_text": "Cheng-Lung Huang and Chieh-Jen Wang. \"A GA-based feature selection and parameters optimizationfor support vector machines.\" Expert Systems with applications 31.2 (2006): 231-240.", "links": null }, "BIBREF12": { "ref_id": "b12", "title": "LIBSVM : a library for support vector machines", "authors": [ { "first": "C-C", "middle": [], "last": "Chang", "suffix": "" }, { "first": "C-J", "middle": [], "last": "Lin", "suffix": "" } ], "year": 2011, "venue": "ACM Transactions on Intelligent Systems and Technology", "volume": "2", "issue": "", "pages": "27--28", "other_ids": {}, "num": null, "urls": [], "raw_text": "C-C Chang and C-J Lin. LIBSVM : a library for support vector machines. ACM Transactions on Intelligent Systems and Technology, 2:27:1--27:27, 2011.Software available at http://www.csie.ntu.edu.tw/~cjlin/libsvm", "links": null }, "BIBREF13": { "ref_id": "b13", "title": "A tutorial on support vector machines for pattern recognition", "authors": [ { "first": "J", "middle": [ "C" ], "last": "Christopher", "suffix": "" }, { "first": "", "middle": [], "last": "Burges", "suffix": "" } ], "year": 1998, "venue": "Data mining and knowledge discovery", "volume": "2", "issue": "", "pages": "121--167", "other_ids": {}, "num": null, "urls": [], "raw_text": "Christopher JC Burges. \"A tutorial on support vector machines for pattern recognition.\" Data mining and knowledge discovery 2.2 (1998): 121-167", "links": null } }, "ref_entries": { "TABREF0": { "num": null, "text": "\u3002\u5728\u672c\u7814\u7a76\u4e2d\u5373\u662f\u5c07\u6240\u6709\u7279\u5fb5\u7684\u96c6\u5408\u8996\u70ba\u67d3\u8272\u9ad4\uff0c \u5404\u500b\u7279\u5fb5\u5373\u70ba\u57fa\u56e0\uff0c\u5229\u7528 GA \u642d\u914d SVM \u5206\u985e\u5668\uff0c\u6700\u5f8c\u6703\u5f97\u5230\u4e00\u4e32 0 \u8207 1 \u7684\u5e8f\u5217\uff0c \u82e5\u70ba 1 \u5247\u4ee3\u8868\u6b64\u7279\u5fb5\u88ab\u9078\u4e2d[12]\uff0c\u53cd\u4e4b\u4ea6\u7136\u3002\u5176\u4e2d GA \u6311\u9078\u65b9\u5f0f\u53ca\u4f7f\u7528\u7684\u53c3\u6578\u5982 GA \u6311\u9078\u5f8c\uff0c\u5728\u7537\u6027\u65b9\u9762\u5171\u6709 259 \u500b\u7279\u5fb5\uff0c\u5176\u4e2d prosodic \u8207\u983b\u57df\u7279\u5fb5\u6709 237 \u500b\uff0c \u4ee5 MFCC\u3001formant \u8207 pitch \u70ba\u4e3b\uff0c\u800c\u975e\u7dda\u6027\u7279\u5fb5 Shannon entopy \u4e4b\u5e73\u5747\u3001\u4e2d\u4f4d\u6578 \u7b49\u5171 5 \u500b\uff0ccurvature index \u4ee5\u767e\u5206\u4f4d\u6578\u70ba\u4e3b\u7684\u7d71\u8a08\u91cf\u5171\u6709 17 \u500b\u3002\u5728\u5973\u6027\u65b9\u9762\uff0c\u6240 \u5f97\u7279\u5fb5\u5171\u6709 247 \u500b\uff0c\u5176\u4e2d prosodic \u8207\u983b\u57df\u7279\u5fb5\u5171\u6709 230 \u500b\uff0c\u4ee5 MFCC\u3001formant \u8207 pitch \u70ba\u4e3b\uff0c\u975e\u7dda\u6027\u7279\u5fb5 Shannon entopy \u4e4b\u5207\u5c3e\u5747\u503c\u8207\u767e\u5206\u4f4d\u6578\u5171 5 \u500b\uff0ccurvature", "html": null, "content": "
\u4e00\u3001 \u7dd2\u8ad6 \u5728\u4eba\u5de5\u667a\u6167\u3001\u6a5f\u5668\u5b78\u7fd2\u8207\u7db2\u8def\u8cc7\u8a0a\u7684\u5feb\u901f\u767c\u5c55\u4e0b\uff0c\u5728\u4e0d\u540c\u9818\u57df\u90fd\u5df2\u7d93\u6709\u8a31\u591a\u4e8b\u60c5 \u53ef\u4ee5\u7531\u6a5f\u5668\u53d6\u4ee3\uff0c\u5982\u6703\u8b70\u5b89\u6392\u3001\u8a9e\u8a00\u5b78\u7fd2\u3001\u8a9e\u97f3\u670d\u52d9\u3001\u65b0\u805e\u64ad\u5831\u3001\u6c7d\u8eca\u99d5\u99db\u7b49\u7b49\uff0c \u8868 \u4e00\u3001\u57fa\u672c\u60c5\u611f\u4e4b\u5b9a\u7fa9 \u5b78\u8005 \u57fa\u672c\u60c5\u611f Arnold Anger, aversion, courage, dejection, desire, despair, dear, hate, hope, love, sadness \u4e8c\u3001 \u7814\u7a76\u65b9\u6cd5 (\u4e00) \u5be6\u9a57\u8cc7\u6599\u5eab 2. \u983b\u8b5c\u7279\u5fb5 \u983b\u57df\u6240\u4f7f\u7528\u4e4b\u7279\u5fb5\uff0c\u7b2c\u4e00\u9805\u70ba\u6885\u723e\u5012\u983b\u8b5c\u4fc2\u6578\uff0c\u914d\u5408\u4eba\u8033\u807d\u89ba\u5c0d\u4e0d\u540c\u983b\u7387\u6709\u4e0d\u540c \u7684\u654f\u611f\u5ea6\u7684\u7279\u6027\uff0c\u63d0\u51fa\u4e86\u9019\u9805\u4fc2\u6578\uff1b\u672c\u7814\u7a76\u6240\u4f7f\u7528\u4e4b pre-emphasis \u4e4b\u9ad8\u901a\u6ffe\u6ce2\u5668 \u4e09\u3001 \u5be6\u9a57\u7d50\u679c \u53bb\u5c0b\u627e\u6700\u9069\u5408\u74b0\u5883\u7684\u57fa\u56e0[11]\u4e0b\u8868\u4e8c[7]\u3002 (\u4e00) FDR \u7d50\u679c
\u4f46\u5982\u679c\u50c5\u50c5\u53ea\u662f\u7531\u6a5f\u5668\u55ae\u65b9\u9762\u63d0\u4f9b\u5236\u5f0f\u5316\u7684\u56de\u61c9\u670d\u52d9\uff0c\u6216\u8a31\u4e0d\u662f\u90a3\u9ebc\u9069\u7576\uff0c\u56e0\u6b64 Ekman, Friesen, Ellsworth Anger, disgust, fear, joy, sadness, surprise \u672c\u7814\u7a76\u7684\u8cc7\u6599\u4f86\u81ea\u65bc\u5fb7\u570b\u67cf\u6797\u8a9e\u97f3\u60c5\u7dd2\u8cc7\u6599\u5eab(Berlin emotion database)[8]\uff0c\u5176\u4e2d \u53c3\u6578\u70ba 0.9\uff0c\u5171\u53d6 13 \u500b\u6885\u723e\u5012\u983b\u8b5c\u4fc2\u6578\u3002\u5171\u632f\u5cf0\u662f\u5c07\u6642\u57df\u8a0a\u865f\u8f49\u70ba\u983b\u57df\u5f8c\uff0c\u53d6\u5176 \u8868 \u4e8c\u3001 GA \u53c3\u6578\u8a2d\u5b9a \u4e0b\u5716\u56db\u70ba FDR \u4e0d\u540c\u7279\u5fb5\u4e4b\u5206\u5e03\u5716\uff0c\u5176\u6a19\u7c64 1 \u81f3 7 \u4ee3\u8868\u4e0d\u540c\u7684\u4e03\u7a2e\u60c5\u7dd2\uff0c\u5716\u56db(a)
\u8b93\u6a5f\u5668\u5075\u6e2c\u5f97\u4eba\u985e\u6240\u8981\u8868\u9054\u7684\u60c5\u7dd2\u8a0a\u606f\uff0c\u63a5\u8457\u7d66\u4e88\u6700\u9069\u7576\u7684\u56de\u61c9\u662f\u4e00\u9805\u91cd\u8981\u7684\u6a5f Fridja Desire, happiness, interest, surprise, wonder, sorrow \u5305\u542b\u4e86\u751f\u6c23(anger)\u3001\u7121\u804a(boredom)\u3001\u53ad\u60e1(disgust)\u3001\u5bb3\u6015(fear)\u3001\u958b\u5fc3(joy)\u3001\u4e2d\u6027 \u5305\u7d61\u7dda(envelope)\u5f8c\u53ef\u5f97\u5230\u4e00\u689d\u8f03\u70ba\u5e73\u6ed1\u7684\u983b\u8b5c\u66f2\u7dda\uff0c\u5176\u4e2d\u6709\u82e5\u5e72\u500b\u9ad8\u9ede\uff0c\u9019\u4e9b Selection technique Roulette wheel \u70ba FDR \u5206\u6790\u5f8c\uff0c\u5c07\u5176\u6700\u5927\u7684\u5169\u500b\u503c\u4ee3\u8868\u7684\u7279\u5fb5\u6240\u756b\u7684\u5206\u5e03\u5716\uff0c\u5716\u56db(b)\u5247\u70ba\u6700\u5c0f
\u5236\u3002\u9019\u4e0d\u50c5\u50c5\u53ef\u4ee5\u589e\u9032\u4eba\u6a5f\u4e92\u52d5\u7684\u6a02\u8da3\uff0c\u4e5f\u53ef\u5728\u4e00\u822c\u5ba2\u670d\u6a5f\u5668\u63d0\u4f9b\u5ba2\u89c0\u8cc7\u8a0a\u5916\uff0c Gray Desire, happiness, interest, surprise, wonder, sorrow (neutral)\u548c\u50b7\u5fc3(sadness)\u5171\u4e03\u7a2e\u60c5\u7dd2\uff0c\u7531\u5341\u4f4d\u5c08\u696d\u6f14\u54e1(\u4e94\u7537\u3001\u4e94\u5973)\u5404\u5225\u6f14\u793a\u4e0a\u8ff0 \u9ad8\u9ede\u8868\u793a\u80fd\u91cf\u96c6\u4e2d\u7684\u4f4d\u7f6e\uff0c\u4e5f\u5c31\u662f\u5171\u632f\u5cf0\uff0c\u53ef\u63cf\u8ff0\u4eba\u985e\u8072\u9053\u4e2d\u7684\u5171\u632f\u60c5\u5f62(\u5982\u4e0b Crossover type Single point crossover \u5169\u500b\u503c\u7684\u7d50\u679c\uff0c\u53ef\u770b\u51fa FDR \u503c\u8d8a\u5927\u4ee3\u8868\u6b64\u7279\u5fb5\u6709\u8f03\u660e\u986f\u7684\u5340\u5206\u6548\u679c\u3002
\u7d66\u4e88\u9069\u5207\u5730\u554f\u5019\u8a71\u8a9e\uff1b\u5728\u667a\u6167\u5bb6\u5ead\u8207\u7167\u8b77\u7cfb\u7d71\u65b9\u9762\uff0c\u82e5\u53ef\u5f97\u77e5\u4f7f\u7528\u8005\u7576\u4e0b\u60c5\u7dd2\u800c \u505a\u51fa\u53cd\u61c9\uff0c\u5982\u5207\u63db\u97f3\u6a02\u3001\u71c8\u5149\u63a7\u5236\u7b49\u7b49\uff0c\u53ef\u4ee5\u63d0\u5347\u4eba\u6a5f\u4e92\u52d5\u7684\u6210\u6548\uff1b\u5176\u4ed6\u50cf\u662f\u5a1b \u6a02\u7522\u54c1\u7684\u4ecb\u9762\u4e5f\u662f\u53ef\u4ee5\u61c9\u7528\u7684\u4e3b\u984c\u3002\u76ee\u524d\u5728\u6a5f\u5668\u8207\u4eba\u7684\u4e92\u52d5\u4e0a\uff0c\u57fa\u672c\u4e0a\u53ef\u5229\u7528\u8996 \u89ba\u8207\u807d\u89ba\u5169\u7a2e\u4eba\u985e\u611f\u5b98\uff0c\u672c\u7814\u7a76\u8457\u91cd\u65bc\u807d\u89ba\u4e4b\u8a9e\u97f3\u60c5\u7dd2\u8fa8\u8b58\u7cfb\u7d71\uff0c\u671f\u671b\u85c9\u7531\u8a9e\u97f3 \u8a0a\u865f\u4f86\u5206\u8fa8\u4f7f\u7528\u8005\u76ee\u524d\u7684\u60c5\u7dd2\uff0c\u9032\u800c\u63d0\u5347\u6e9d\u901a\u6548\u679c\u3002 \u5c0d\u65bc\u60c5\u7dd2\u7684\u63cf\u8ff0\u65b9\u5f0f\u5927\u81f4\u53ef\u5206\u70ba\u96e2\u6563\u8207\u7dad\u5ea6\u5169\u7a2e\u5f62\u5f0f\uff0c\u524d\u8005\u5373\u70ba\u65e5\u5e38\u751f\u6d3b\u6240\u4f7f\u7528 \u4e4b\u8a5e\u5f59\uff0c\u5982\u958b\u5fc3\u3001\u751f\u6c23\u3001\u60b2\u50b7\u7b49\uff0c\u5728\u5982\u6b64\u5927\u91cf\u4e4b\u60c5\u611f\u8a5e\u5f59\u4e2d\uff0c\u4e00\u822c\u8a8d\u70ba\u80fd\u5920\u70ba\u4eba \u985e\u8207\u5177\u6709\u793e\u6703\u6027\u4e4b\u54fa\u4e73\u52d5\u7269\u6240\u5171\u6709\u60c5\u611f\u7a31\u70ba\u57fa\u672c\u60c5\u611f\uff0c\u4e0d\u540c\u5b78\u8005\u5c0d\u65bc\u57fa\u672c\u60c5\u611f\u7684 \u5b9a\u7fa9\u4e5f\u4e0d\u76f8\u540c\uff0c\u5176\u4e2d\u4ee5 Ekman \u63d0\u51fa\u4e4b\u516d\u5927\u57fa\u672c\u60c5\u611f\u8f03\u70ba\u5ee3\u6cdb\u88ab\u4f7f\u7528\uff0c\u7576\u7136\u4ea6\u6709 \u8a31\u591a\u4f9d\u6b64\u767c\u5c55\u6216\u5176\u4ed6\u7406\u8ad6\u800c\u5f62\u6210\u7684\u57fa\u672c\u60c5\u7dd2\uff0c\u5982\u4e0b\u8868\u4e00[1]\uff1b\u5f8c\u8005\u5247\u5c07\u60c5\u611f\u72c0\u614b \u63cf\u8ff0\u65bc\u6fc0\u6d3b\u5ea6-\u6548\u50f9\u60c5\u611f\u7a7a\u9593(arousal-valence emotional space)\u6216\u662f\u6fc0\u52f5-\u6548\u50f9-\u63a7 \u5236\u7a7a\u9593(activation -valence -dominance space)\u4e2d\uff0c\u5176\u4e2d\u6bcf\u4e00\u500b\u7dad\u5ea6\u5c0d\u61c9\u8457\u5fc3\u7406\u5b78\u7684 \u5c6c\u6027[2\u30013]\u3002\u57fa\u672c\u4e0a\uff0c\u900f\u904e\u8072\u97f3\u4f86\u50b3\u905e\u60c5\u7dd2\u4e0a\u5927\u81f4\u53ef\u5206\u70ba\u5169\u500b\u65b9\u5411\uff0c\u4e00\u70ba\u900f\u904e\u8a9e \u610f\uff0c\u5373\u7531\u5b57\u9762\u4e0a\u7684\u610f\u601d\uff1b\u53e6\u5916\u662f\u85c9\u7531\u8a9e\u8abf\u4f86\u50b3\u905e\u60c5\u7dd2\u3002\u800c\u5728\u672c\u7814\u7a76\u4e2d\u5247\u63a1\u7528\u4e86\u96e2 \u6563\u60c5\u7dd2\u5206\u985e\u53ca\u900f\u904e\u8a9e\u8abf\u4f86\u64f7\u53d6\u7279\u5fb5\uff0c\u9032\u800c\u4f5c\u60c5\u7dd2\u5206\u985e\u5224\u65b7\u3002 \u904e\u53bb\u6587\u737b\u4e2d\uff0cMoataz El Ayadi \u7b49\u4eba[4]\u63d0\u4f9b\u4e0d\u540c\u8a9e\u6599\u5eab\u6536\u96c6\u65b9\u5f0f\u4e4b\u8cc7\u8a0a\u53ca\u8a31\u591a\u8a9e\u97f3 \u8a0a\u865f\u7279\u5fb5\u4e4b\u8a08\u7b97\u65b9\u5f0f\u8207\u5206\u985e\u65b9\u6cd5\uff1bSiqing Wu \u7b49[5]\u5229\u7528\u8abf\u8b8a\u983b\u8b5c\u7279\u5fb5(MSFs)\u8207\u4e0d \u540c\u7279\u5fb5\u7d44\u5408\u9032\u884c\u60c5\u7dd2\u5206\u985e\uff0c\u5176 \u6700\u4f73\u6e96\u78ba\u7387 91.6%\u70ba MSFs \u8207\u8072\u97fb(prosodic)\u7279\u5fb5\u7684 Anger, contempt, disgust, distress, fear, guilt, interest, \u4e03\u7a2e\u60c5\u7dd2\u5c0d\u61c9\u7684\u53e5\u5b50\u6240\u7d44\u6210\uff0c\u5171\u6709 535 \u53e5\u8a9e\u97f3\u8a0a\u865f\u3002 \u5716\u4e8c)\u3002\u672c\u7814\u7a76\u5229\u7528\u5feb\u901f\u5085\u7acb\u8449\u8f49\u63db(FFT)\u53ca linear predictive coding(LPC)\u65b9\u5f0f\u53d6\u5f97 Population size 50 Izard joy, shame, surprise James \u7b2c 1 \u5230\u7b2c 3 \u500b\u5171\u632f\u5cf0(F1~F3)\u7684\u983b\u7387\u503c\u53ca\u5176\u983b\u5bec\u3002 Crossover rate 0.9 (\u4e8c) \u7279\u5fb5\u64f7\u53d6 Mutation rate 0.001 Fear, grief, love, rage Fear, disgust, elation, fear, subjection, \u5c07\u8a9e\u97f3\u8a0a\u865f\u9032\u884c\u97f3\u6846(frame)\u7684\u5207\u5272\uff0c\u901a\u5e38\u8996\u7a97\u9577\u5ea6\u70ba 20~40ms\uff0c\u7528\u4f86\u8a08\u7b97\u7279\u5fb5\u53c3 \u5716 \u4e09\u3001 Curvature index \u8a08\u7b97\u7d50\u679c Iteration number 200 McDougall tender-emotion, wonder Mower Pain, pleasure Oatley, Johnson-Laird \u6578\uff0c\u800c\u70ba\u4e86\u8b93\u7279\u5fb5\u8b8a\u5316\u6709\u5ef6\u7e8c\u6027\uff0c\u6703\u5c07\u90e8\u5206\u8996\u7a97\u91cd\u758a(overlap)\uff0c\u672c\u7814\u7a76\u6240\u4f7f\u7528\u4e4b 4. \u7d71\u8a08\u503c \u8996\u7a97\u9577\u5ea6\u70ba 32ms\uff0c\u91cd\u758a\u90e8\u5206\u70ba 16ms\u3002\u64f7\u53d6\u7684\u7279\u5fb5\u5206\u70ba\u5169\u90e8\u5206\uff0c\u4e00\u70ba\u50b3\u7d71\u4f7f\u7528\u4e4b \u8072\u97fb\u548c\u983b\u8b5c\u7279\u5fb5\uff0c\u53e6\u4e00\u90e8\u5206\u5247\u662f\u975e\u7dda\u6027\u52d5\u614b\u7279\u5fb5 Shannon entropy \u548c curvature \u8a08\u7b97\u8a9e\u97f3\u8a0a\u865f\u6bcf\u500b\u97f3\u6846\u7684\u4e0a\u8ff0\u7279\u5fb5\u503c\u5f8c\u9032\u884c\u7d71\u8a08\uff0c\u5176\u7d71\u8a08\u91cf\u5305\u542b\u6700\u5c0f\u503c(min)\u3001 (\u56db) \u5206\u985e\u65b9\u5f0f \u5716 \u516d\u3001\u7537\u6027\u5206\u985e\u7d50\u679c \u5716 \u4e94\u3001 \u5973\u6027\u5206\u985e\u7d50\u679c Anger, disgust, anxiety, happiness, sadness Panksepp Panksepp Anger, disgust, anxiety, happiness, sadness Plutchik Acceptance, anger, anticipation, disgust, joy, fear, index\u3002 1. \u8072\u97fb\u7279\u5fb5 \u5728\u8072\u97fb\u7279\u5fb5\u4e2d\uff0c\u6536\u96c6\u4e86\u97f3\u9ad8\u3001\u80fd\u91cf\u3001\u904e\u96f6\u7387(zero crossing rate,ZCR)\u3001TEO(Teager \u6700\u5927\u503c(max)\u3001\u6700\u5927\u8207\u6700\u5c0f\u503c\u7684\u5dee(range)\u3001\u5e73\u5747(mean)\u3001\u4e2d\u4f4d\u6578(median)\u3001\u5207\u5c3e\u5747 \u5728\u7279\u5fb5\u64f7\u53d6\u524d\uff0c\u5df2\u5c07\u8cc7\u6599\u4ee5 80%\u8207 20%\u7684\u6bd4\u4f8b\u5206\u70ba\u8a13\u7df4\u8cc7\u6599\u96c6(training data set) (\u4e0a\u5716\u70ba\u50b3\u7d71\u7279\u5fb5\uff0c\u4e0b\u5716\u70ba\u65b0\u589e\u975e\u7dda\u6027\u7279\u5fb5) (\u4e0a\u5716\u70ba\u50b3\u7d71\u7279\u5fb5\uff0c\u4e0b\u5716\u70ba\u65b0\u589e\u975e\u7dda\u6027\u7279\u5fb5) \u5716 \u4e8c\u3001Formant \u7d50\u679c 3. \u975e\u7dda\u6027\u52d5\u614b\u7279\u5fb5 \u503c(trimmed mean)\u4e4b 10%\u8207 25%\u3001\u7b2c 1\u30015\u300110\u300125\u300175\u300190\u300195\u300199 \u7684\u767e\u5206\u4f4d \u6578(percentile)\u3001\u56db\u5206\u5dee(interquartile range)\u3001\u5e73\u5747\u5dee(average deviation)\u3001\u6a19\u6e96\u5dee (standard deviation)\u3001\u504f\u614b(skewness)\u548c\u5cf0\u5ea6(kurtosis)\u5171 20 \u9805\u3002\u53e6\u5916\u4e5f\u8a08\u7b97\u76f8\u9130\u5169 \u8207\u9a57\u8b49\u8cc7\u6599\u96c6(validation data set)\uff0c\u9a57\u8b49\u8cc7\u6599\u96c6\u5167\u6240\u6709\u8cc7\u6599\u7686\u4e0d\u6703\u7d93\u904e\u6311\u9078\u8207\u5206\u985e\uff0c (a) (b) \u800c\u662f\u4f5c\u70ba\u8a13\u7df4\u6a21\u578b\u597d\u58de\u7684\u5224\u65b7\u4f9d\u64da\uff0c\u672c\u7814\u7a76\u6240\u4f7f\u7528\u4e4b\u5206\u985e\u5668 SVM\uff0c\u63a1\u7528\u7684 toolbox \u70ba LibSVM [13]\u3002 \u56db\u3001 \u7d50\u8ad6 \u5716 \u56db\u3001(a) FDR \u6700\u5927\u5169\u503c\u7279\u5fb5\u5206\u5e03\u5716\uff0c(b)FDR \u6700\u5c0f\u5169\u503c\u7279\u5fb5\u5206\u5e03\u5716 sadness, surprise energy operator)\u7b49\u5e38\u898b\u8a9e\u97f3\u5206\u6790\u7279\u5fb5\u3002\u97f3\u9ad8\u64f7\u53d6\u65b9\u5f0f\u662f\u4f7f\u7528 ACF(auto-correlation \u590f\u8fb2\u71b5\u5728\u8cc7\u8a0a\u7406\u8ad6\u4e2d\u626e\u6f14\u4e86\u5f88\u91cd\u8981\u7684\u89d2\u8272\uff0c\u9664\u4e86\u53ef\u7528\u4f86\u4f5c\u70ba\u8cc7\u8a0a\u91cf\u7684\u91cf\u6e2c\u5916\uff0c\u540c \u97f3\u6846\u4e4b\u4e00\u968e\u8207\u4e8c\u968e\u5012\u6578\u4e4b\u7d71\u8a08\u91cf\uff0c\u4ee5\u8868\u793a\u5169\u97f3\u6846\u9593\u7684\u8b8a\u5316\u7a0b\u5ea6\uff0c\u6700\u5f8c\u5c07\u6240\u6709\u7d71\u8a08 \u672c\u7814\u7a76\u4ee5\u4e00\u822c\u5e38\u7528\u4e4b\u8a9e\u97f3\u7279\u5fb5\u97f3\u9ad8\u3001\u5171\u632f\u5cf0\u3001\u80fd\u91cf\u4ee5\u53ca\u6885\u723e\u5012\u983b\u8b5c\u4fc2\u6578\u70ba\u57fa\u790e\uff0c Tomkins Tomkins Anger, interest, contempt, disgust, distress, fear, joy, shame, surprise Watson Fear, love rage Weiner, Graham Happiness, sadness function)\uff0c\u4f46\u70ba\u4e86\u907f\u514d ACF \u7684\u503c\u4ecb\u65bc\u4e00\u500b\u4e0d\u5b9a\u7684\u5340\u9593\uff0c\u5c07\u5176\u6b63\u898f\u5316\u81f3 1 \u8207-1 \u4e4b \u9593\u5f8c\uff0c\u518d\u642d\u914d\u97f3\u91cf\u95be\u503c\u5224\u65b7\u97f3\u9ad8\uff0c\u5373\u5f97NACF(\u03c4) = 2 \u2211 s(i)s(i+\u03c4) \u2211 s 2 (i)+ \u2211 s 2 (i+\u03c4) \u3002\u904e\u96f6\u7387\u5373\u70ba\u8a0a \u865f\u904e\u96f6\u9ede\u7684\u6b21\u6578\uff0c\u4e00\u822c\u800c\u8a00\u5176\u503c\u5728\u6709\u8a9e\u97f3\u7684\u6642\u5019\u6703\u6bd4\u5b89\u975c\u6216\u74b0\u5883\u96dc\u8a0a\u8f03\u5927\u6642\u4f4e\uff0c \u56e0\u6b64\u672c\u7814\u7a76\u63a1\u7528\u6b64\u65b9\u6cd5\u642d\u914d\u97f3\u91cf\u4f86\u5224\u65b7 voice activity ratio\uff0cvoice activity ratio \u5373 \u70ba\u4e00\u6bb5\u8a0a\u865f\u5167\u6709\u8a9e\u97f3\u8207\u7121\u8a9e\u97f3\u7684\u6bd4\u4f8b(\u5982\u4e0b\u5716\u4e00)\u3002TEO \u5247\u662f\u5728\u9084\u539f\u8072\u97f3\u7d93\u904e\u6c23\u7ba1 \u53ca\u4eba\u7684\u8154\u9ad4\u4f5c\u7528\u5f8c\u6240\u7522\u751f\u7684\u8a9e\u97f3\u8a0a\u865f\uff0cTEO(s i ) = s i 2 \u2212 s i\u22121 s i+1 \uff0c\u4e0a\u8ff0\u516c\u5f0f\u5167\u4e4b s \u5373\u70ba\u4e00\u500b\u97f3\u6846\u5167\u7684\u539f\u59cb\u8a0a\u865f\uff0ci \u8868\u793a\u7b2c i \u9ede\u8a0a\u865f\u3002 \u6642\u4e5f\u662f\u5c0d\u67d0\u500b\u7cfb\u7d71\u4e4b\u4e0d\u78ba\u5b9a\u6027\u6216\u6df7\u4e82\u7a0b\u5ea6\u7684\u5ea6\u91cf\u65b9\u6cd5\uff0c\u82e5\u71b5\u503c\u8d8a\u9ad8\u5247\u7cfb\u7d71\u7684\u4e0d\u78ba \u5b9a\u6027(uncertainty)\u8d8a\u9ad8\uff0c\u53cd\u4e4b\u4ea6\u7136\u3002\u96a8\u6a5f\u8b8a\u6578 \u7684\u590f\u8fb2\u71b5\u53ef\u5b9a\u7fa9\u70ba H( ) = \u2212 \u2211 ( ) log ( ) \u2208 \uff0c \u5176\u4e2d ( ) = { = }, \u2208 \u3002\u4f7f\u7528\u4e0d\u540c\u57fa\u5e95\u6703\u6709\u4e00\u8f49\u63db\u5e38\u6578\u7684\u5dee\u7570\u3002 \u66f2\u7387\u6307\u6a19[9]\u662f\u4e00\u52d5\u614b\u7cfb\u7d71\u7684\u6307\u6a19\uff0c\u66f2\u7387\u6307\u6a19\u4e4b\u5b9a\u7fa9\u5982\u4e0b\uff0c\u5c0d\u65bc \u7dad\u7a7a\u9593\u66f2\u7dda (t) \u2208 \u211d \u53ef\u5f97 \u2212 1\u500b\u9ad8\u7dad\u5ea6\u66f2\u7387\u03ba i , 1 \u2264 i \u2264 \u2212 1\uff0c\u5247\u66f2\u7387\u6307\u6a19\u70ba \u91cf\u7576\u4f5c\u8a9e\u97f3\u8a0a\u865f\u4e4b\u7279\u5fb5\u9032\u884c\u6311\u9078\u8207\u5206\u985e\u3002 (\u4e09) \u7279\u5fb5\u6311\u9078 \u7279\u5fb5\u9078\u53d6\u7684\u76ee\u6a19\u662f\u8981\u5f9e\u539f\u6709\u7684\u7279\u5fb5\u96c6\u5408\u4e2d\u6311\u9078\u51fa\u9451\u5225\u80fd\u529b\u8f03\u597d\u7684\u7279\u5fb5\uff0c\u4f7f\u5176\u8fa8\u8b58 \u7387\u80fd\u5920\u9054\u5230\u6700\u9ad8\u503c\uff0c\u4e0d\u4f46\u80fd\u5920\u7c21\u5316\u5206\u985e\u5668\u7684\u8a08\u7b97\uff0c\u4e26\u53ef\u85c9\u6b64\u4e86\u89e3\u5206\u985e\u554f\u984c\u95dc\u4fc2\u3002 \u7279\u5fb5\u6311\u9078\u6642\u4f7f\u7528\u4e86 10 \u6298\u4ea4\u53c9\u9a57\u8b49(10-fold cross validation)\uff0c\u907f\u514d\u5c0d\u55ae\u4e00\u8cc7\u6599\u5f62\u6210 over-fitting\u3002 SVM \u662f\u4e00\u7a2e\u6a5f\u5668\u5b78\u7fd2\u7684\u6f14\u7b97\u6cd5\uff0c\u76ee\u7684\u662f\u70ba\u4e86\u5efa\u7acb\u4e00\u500b\u6a21\u578b\u4ee5\u8fa8\u5225\u4e0d\u540c\u8cc7\u6599\u7684\u985e \u52a0\u5165\u4e86\u975e\u7dda\u6027\u7279\u5fb5 Shannon entropy \u548c curvature index\uff0c\u7d93\u7531\u7279\u5fb5\u64f7\u53d6\u3001\u7279\u5fb5\u6311\u9078 (\u4e8c) GA \u6311\u9078\u7d50\u679c \u5230\u6700\u5f8c\u5206\u985e\u7684\u65b9\u5f0f\u5efa\u7acb\u8a9e\u97f3\u60c5\u7dd2\u8fa8\u8b58\u6a21\u578b\u3002\u4ee5\u67cf\u6797\u8a9e\u97f3\u60c5\u7dd2\u8cc7\u6599\u5eab\u505a\u70ba\u5206\u6790\u5c0d\u8c61\uff0c \u5225\uff0c\u5229\u7528 SVM \u642d\u914d\u6838\u65b9\u6cd5(kernel method)\u53ef\u4ee5\u6709\u6548\u7387\u5730\u5c07\u539f\u59cb\u8cc7\u6599\u8f49\u63db\u5230\u9ad8\u7dad\u5ea6 \u7684\u7a7a\u9593\uff0c\u4e26\u5728\u8a13\u7df4\u8cc7\u6599\u96c6\u4e2d\u627e\u51fa\u9918\u88d5(margin)\u6700\u5927\u7684\u8d85\u5e73\u9762(hyper-plane)\uff0c\u6b64 hyper-plane \u5c07\u6703\u662f\u6e2c\u8a66\u8cc7\u6599\u7684\u5206\u985e\u4f9d\u64da\uff0c\u900f\u904e\u6b64\u65b9\u6cd5\u6211\u5011\u53ef\u5f97\u5230\u4e00\u500b\u6e96\u78ba\u7387\u9ad8\u4e14 \u5177\u6709\u9ad8\u6297\u96dc\u8a0a\u529f\u80fd\u7684\u5206\u985e\u6a21\u578b\uff0c\u53e6\u5916\u76f8\u8f03\u65bc\u5176\u4ed6\u6a5f\u5668\u5b78\u7fd2\u800c\u8a00\uff0c\u5c0d\u65bc\u6578\u91cf\u8f03\u5c11\u7684 \u8cc7\u6599\u5176\u932f\u8aa4\u7387\u53ca\u8907\u96dc\u6027\u53ef\u88ab\u6700\u5c0f\u5316[14]\u3002 \u672a\u52a0\u5165\u975e\u7dda\u6027\u7279\u5fb5\u91cf\uff0c\u6240\u5f97\u7537\u6027\u53ca\u5973\u6027\u4e4b\u60c5\u7dd2\u8fa8\u8b58\u7387\u5206\u5225\u70ba 84.44%\u53ca 84.48%\uff1b \u52a0\u5165\u975e\u7dda\u6027\u7279\u5fb5\u91cf\u4e4b\u5f8c\uff0c\u7537\u6027\u8fa8\u8b58\u7387\u63d0\u9ad8\u81f3 88.89%\uff0c\u5973\u6027\u5247\u63d0\u9ad8\u81f3 86.21%\u3002 \u91dd\u5c0d\u5404\u5225\u60c5\u7dd2\u8fa8\u8b58\u6539\u9032\u7684\u7d30\u90e8\u7d50\u679c\u65b9\u9762\uff0c\u53ef\u7531 Confusion matrix(\u5716\u4e94\u3001\u5716\u516d)\u5f97\u77e5\uff0c \u5728\u52a0\u5165\u975e\u7dda\u6027\u7279\u5fb5\u91cf\u5f8c\uff0c\u5973\u6027\u65b9\u9762\u5247\u56e0\u70ba\u8aa4\u5224\u70ba\u5bb3\u6015\u4e4b\u958b\u5fc3\u60c5\u7dd2\u6709\u90e8\u5206\u88ab\u6539\u6b63\uff0c \u7d93 index \u5247\u4ee5\u767e\u5206\u4f4d\u6578\u70ba\u4e3b\u7684\u7d71\u8a08\u91cf\u5171 12 \u500b\u3002 \u4f7f\u6e96\u78ba\u7387\u7531 62.5%\u63d0\u5347\u70ba 75%\uff1b\u7537\u6027\u65b9\u9762\u7531\u65bc\u539f\u672c\u88ab\u8aa4\u5224\u70ba\u7121\u804a\u7684\u4e2d\u6027\u60c5\u7dd2\u5df2\u5224
\u7d44\u5408\u6cd5\uff1bPatricia Henr\u00edquez[6]\u7b49\u5229\u7528\u975e\u7dda\u6027\u52d5\u614b\u7279\u5fb5\u9032\u884c\u8a9e\u97f3\u60c5\u7dd2\u8fa8\u8b58\u7814\u7a76\uff0c\u6e96 \u78ba\u7387\u6700\u9ad8\u53ef\u9054 80.75%\uff1bAli Shahzadi \u7b49[7]\u4ee5\u8072\u97fb\u7279\u5fb5\u3001\u983b\u8b5c\u7279\u5fb5\u8207\u975e\u7dda\u6027\u52d5\u614b\u7279 \u5fb5\u4f9d\u4e0d\u540c\u7d44\u5408\u9032\u884c\u7814\u7a76\uff0c\u5176\u6e96\u78ba\u7387\u6700\u9ad8\u70ba\u7537\u6027 85.9%\uff0c\u5973\u6027\u70ba 82.72%\u3002\u672c\u7814\u7a76 \u039a = lim \u7531\u4e0a\u5f0f\u53ef\u77e5\uff0c\u66f2\u7387\u6307\u6a19\u662f\u85c9\u7531\u52d5\u614b\u5e73\u5747\u7684\u65b9\u5f0f\u4f86\u63cf\u8ff0\uff0c\u5176\u529f\u7528\u5728\u65bc\u7cfb\u7d71\u51fa\u73fe\u7d50\u69cb algorithm, GA)\u9032\u884c\u7279\u5fb5\u6311\u9078\u3002\u4f9d\u64da\u8cbb\u96ea\u5224\u5225\u5206\u6790\u7684\u6982\u5ff5\uff0c\u5206\u5c6c\u4e8c\u500b\u985e\u5225\u7684\u7279\u5fb5 \u6bd4\u8f03\u4e0d\u540c\u6027\u5225\u4f7f\u7528\u50b3\u7d71 prosodic \u8207\u983b\u8b5c\u7279\u5fb5\u548c\u52a0\u5165\u975e\u7dda\u6027\u7279\u5fb5\u5f8c\u7684\u6df7\u6dc6\u77e9\u9663 \u2192\u221e \u222b ( ) 0 , 1 \u2264 \u2264 \u2212 1\u3002 \u672c\u7814\u7a76\u5229\u7528\u4e86\u8cbb\u96ea\u9451\u5225\u6bd4(Fisher discriminate ratio, FDR)\u8207\u57fa\u56e0\u6f14\u7b97\u6cd5(genetic \u65b7\u6b63\u78ba\uff0c\u4f7f\u5f97\u4e2d\u6027\u6e96\u78ba\u7387\u7531 85.71%\u63d0\u5347\u70ba 100%\uff1b\u800c\u53ad\u60e1\u5c07\u539f\u672c\u8aa4\u5224\u70ba\u5bb3\u6015\u7684\u60c5 (\u4e09) SVM \u5206\u985e\u7d50\u679c \u6cc1\u6539\u6b63\uff0c\u81f4\u4f7f\u5176\u6e96\u78ba\u7387\u7531 50%\u5347\u81f3 100%\u3002
\u7684\u76ee\u6a19\u662f\u900f\u904e\u5206\u6790\u8a9e\u97f3\u4f86\u8fa8\u8b58\u60c5\u7dd2\uff0c\u4ee5\u904e\u53bb\u5b78\u8005\u4e4b\u7814\u7a76\u70ba\u57fa\u790e\uff0c\u5229\u7528\u8a9e\u97f3\u8a0a\u865f\u64f7 \u8b8a\u5316\u6642\uff0c\u53ef\u4ee5\u5728\u6307\u6a19\u4e0a\u51fa\u73fe\u76f8\u61c9\u8b8a\u5316\uff0c\u662f\u4ee5\u543e\u4eba\u9810\u671f\uff0c\u7576\u4e0d\u540c\u60c5\u7dd2\u8b8a\u5316\u8868\u73fe\u5728\u8a9e \u5176\u7d44\u5167\u5dee\u8ddd\u8d8a\u5c0f\uff0c\u7d44\u9593\u5dee\u8ddd\u8d8a\u5927\uff0c\u53ef\u7372\u5f97\u8d8a\u597d\u7684\u5206\u985e\u6548\u679c\u3002\u591a\u7d44\u985e\u5225\u4e4b FDR \u8a08 (confusion matrix)\uff0c\u4e0b\u5716\u4e94\u70ba\u5973\u6027\uff0c\u4f7f\u7528\u50b3\u7d71\u7279\u5fb5\u6e96\u78ba\u7387\u70ba 84.48%\uff0c\u52a0\u5165\u975e\u7dda\u6027 \u53e6\u5916\uff0c\u56e0\u76ee\u524d\u6240\u4f7f\u7528\u4e4b\u8cc7\u6599\u70ba\u5fb7\u6587\uff0c\u5c0d\u65bc\u4e0d\u540c\u8a9e\u8a00\u53ca\u6587\u5316\u7684\u5728\u8a9e\u97f3\u60c5\u7dd2\u5f71\u97ff\u7684\u5dee
\u53d6\u7279\u5fb5\u91cf\uff0c\u518d\u4ee5\u6311\u9078\u5f8c\u7684\u7279\u5fb5\u91cf\u4f5c\u70ba\u652f\u6301\u5411\u91cf\u6a5f(support vector machine, SVM)\u4e2d \u97f3\u8a0a\u865f\u6642\uff0c\u5176\u5c0d\u61c9\u7684\u66f2\u7387\u6307\u6a19\u4e5f\u6703\u6709\u6240\u4e0d\u540c\u3002\u8a08\u7b97\u66f2\u7387\u6307\u6a19\u524d\uff0c\u9700\u8981\u904b\u7528\u76f8\u7a7a\u9593 \u7b97\u65b9\u5f0f\u5982\u4e0b[10] \u7279\u5fb5\u5f8c\u63d0\u5347\u81f3 86.21%\uff1b\u5716\u516d\u70ba\u7537\u6027\uff0c\u4f7f\u7528\u50b3\u7d71\u7279\u5fb5\u6e96\u78ba\u7387\u70ba 84.44%\uff0c\u52a0\u5165\u975e\u7dda \u7570\u4e26\u672a\u5728\u7814\u7a76\u4e2d\u63a2\u8a0e\uff0c\u56e0\u6b64\u6709\u8a08\u756b\u5efa\u7acb\u4e2d\u6587\u8a9e\u97f3\u60c5\u7dd2\u8cc7\u6599\u5eab\uff0c\u85c9\u4ee5\u9a57\u8b49\u672c\u7814\u7a76\u65b9
\u7684\u8a13\u7df4\u8cc7\u6599\uff0c\u85c9\u6b64\u8a13\u7df4\u51fa\u5206\u985e\u6a21\u578b\uff0c\u7d50\u679c\u8b49\u660e\u5728\u4e00\u822c\u5e38\ufa0a\u8a9e\u97f3\u7279\u5fb5\u5982\u97f3\u9ad8(pitch)\u3001 \u80fd\u91cf(energy)\u3001\u5171\u632f\u5cf0(formant)\u3001\u6885\u723e\u5012\u983b\u8b5c\u4fc2\u6578(Mel-scale Frequency Cepstral Coefficients, MFCC)\uff0c\u984d\u5916\u52a0\u5165\u4e86\u590f\u8fb2\u71b5(Shannon entropy)\u548c\u66f2\u7387\u6307\u6a19\u5169\u9805\u975e\u7dda\u6027 \u5716 \u4e00\u3001 Voice activity detection\uff0c\u7da0\u8272\u7dda\u70ba\u8d77\u59cb\u4f4d\u7f6e\uff0c\u7d05\u8272\u7dda\u70ba\u7d50\u675f \u91cd\u69cb\u7684\u6280\u8853\u5c07\u8a9e\u97f3\u8a0a\u865f\u91cd\u69cb\u5230\u9ad8\u7dad\u5ea6\u7a7a\u9593\u4e0a\uff0c\u672c\u7814\u7a76\u4e2d\u91cd\u69cb\u7dad\u5ea6 = 3\uff0c\u4e14\u53ea\u6709 K 1 \u5728\u7279\u5fb5\u6311\u9078\u904e\u7a0b\u4e2d\u88ab\u9078\u4e2d\u3002 FDR(u) = 2 C(C \u2212 1) \u2211 \u2211 \u03c3 c 1 c 2 c 1 2 + \u03c3 c 2 2 , 1 \u2264 c 1 < c 2 \u2264 C\uff0c (\u00b5 c 1 ,\u00b5 \u2212 \u00b5 c 2 ,\u00b5 ) 2 \u6027\u7279\u5fb5\u5f8c\u63d0\u5347\u81f3 88.89%\u3002 \u6cd5\u5c0d\u65bc\u4e2d\u6587\u8a9e\u97f3\u60c5\u7dd2\u8fa8\u8b58\u4e4b\u53ef\u884c\u6027\u3002
\u7279\u5fb5\u6709\u63d0\u5347\u8a9e\u97f3\u60c5\u7dd2\u8fa8\u8b58\u4e4b\u6548\u7528\u3002 \u5229\u7528 FDR \u5c07\u4e0d\u9069\u7528\u4e4b\u7279\u5fb5\u6392\u9664\u5f8c\uff0c\u518d\u7d93\u7531 GA \u6311\u51fa\u6700\u5f8c\u8fa8\u5225\u6240\u4f7f\u7528\u7684\u7279\u5fb5\uff0cGA
\u662f\u4eba\u985e\u4f9d\u7167\u751f\u7269\u5b78\u4e2d \u300c\u9069\u8005\u751f\u5b58\uff0c\u4e0d\u9069\u8005\u6dd8\u6c70\u300d \u7684\u89c0\u5ff5\u6240\u767c\u5c55\u51fa\u4f86\u7684\u4e00\u7a2e\u6f14\u7b97\u6cd5\uff0c
\u5229\u7528\u9078\u64c7(selection)\u3001\u8907\u88fd(reinsertion)\u3001\u4ea4\u914d(cross-over)\u3001\u7a81\u8b8a(mutation)\u7b49\u6b65\u9a5f
", "type_str": "table" } } } }