| { |
| "paper_id": "O17-1024", |
| "header": { |
| "generated_with": "S2ORC 1.0.0", |
| "date_generated": "2023-01-19T07:59:46.276401Z" |
| }, |
| "title": "Using Teacher-Student Model For Emotional Speech Recognition", |
| "authors": [ |
| { |
| "first": "\u856d\u4f2f\u744b", |
| "middle": [], |
| "last": "Po-Wei", |
| "suffix": "", |
| "affiliation": { |
| "laboratory": "", |
| "institution": "Information Engineering National Sun Yat-sen University", |
| "location": {} |
| }, |
| "email": "" |
| }, |
| { |
| "first": "Hsiao", |
| "middle": [], |
| "last": "\u570b\u7acb\u4e2d\u5c71\u5927\u5b78\u8cc7\u8a0a\u5de5\u7a0b\u5b78\u7cfb", |
| "suffix": "", |
| "affiliation": { |
| "laboratory": "", |
| "institution": "Information Engineering National Sun Yat-sen University", |
| "location": {} |
| }, |
| "email": "" |
| }, |
| { |
| "first": "Po-Chen", |
| "middle": [], |
| "last": "\u8b1d\u535a\u4e1e", |
| "suffix": "", |
| "affiliation": {}, |
| "email": "" |
| }, |
| { |
| "first": "\u570b\u7acb\u4e2d\u5c71\u5927\u5b78\u8cc7\u8a0a\u5de5\u7a0b\u5b78\u7cfb", |
| "middle": [], |
| "last": "Hsieh", |
| "suffix": "", |
| "affiliation": {}, |
| "email": "" |
| }, |
| { |
| "first": "Chia-Ping", |
| "middle": [], |
| "last": "\u9673\u5609\u5e73", |
| "suffix": "", |
| "affiliation": {}, |
| "email": "" |
| }, |
| { |
| "first": "\u570b\u7acb\u4e2d\u5c71\u5927\u5b78\u8cc7\u8a0a\u5de5\u7a0b\u5b78\u7cfb", |
| "middle": [], |
| "last": "Chen", |
| "suffix": "", |
| "affiliation": {}, |
| "email": "cpchen@mail.cse.nsysu.edu.tw" |
| } |
| ], |
| "year": "", |
| "venue": null, |
| "identifiers": {}, |
| "abstract": "", |
| "pdf_parse": { |
| "paper_id": "O17-1024", |
| "_pdf_hash": "", |
| "abstract": [], |
| "body_text": [ |
| { |
| "text": "\u5f97\u8f38\u51fa\u503c\u4f4d\u65bc\u5340\u9593[0,1]\u51fd\u6578\u5982\u4e0b\u3002 f (x) = 1 1 + e \u2212x", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "(1)", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "EQUATION", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [ |
| { |
| "start": 0, |
| "end": 8, |
| "text": "EQUATION", |
| "ref_id": "EQREF", |
| "raw_str": "\u6703\u4ee5Cross entropy(2)\u4f5c\u70ba\u8a08\u7b97\u8aa4\u5dee\u7684\u65b9\u6cd5 \u2212 K \u2211 k=1 target k \u00d7 log(predict k ) (2) \u52a0\u5165\u6b0a\u91cd\u4e4b\u6700\u4f73\u5316\u53c3\u6578\u66f4\u65b0\u4ee5\u4e0b\u5217\u516c\u5f0f\u505a\u70ba\u8868\u793a \u2212r ik \u00d7 K \u2211 k=1 target k \u00d7 log(predict k )", |
| "eq_num": "(3)" |
| } |
| ], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "\u6700\u5f8c\u4f7f\u7528 Softmax function(4)\u5c07\u5404\u985e\u8f38\u51fa\u7d50\u679c\u8f49\u63db\u70ba\u6a5f\u7387\u3002", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "EQUATION", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [ |
| { |
| "start": 0, |
| "end": 8, |
| "text": "EQUATION", |
| "ref_id": "EQREF", |
| "raw_str": "\u03c3 (y i ) = e y i \u2211 j e y j", |
| "eq_num": "(4)" |
| } |
| ], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "\u5728\u9032\u884c\u5411\u5f8c\u50b3\u905e\u6642:\u8a08\u7b97\u8f38\u51fa\u5c64(5)\u53ca\u96b1\u85cf\u5c64(6)\u7684\u8aa4\u5dee\u503c", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "EQUATION", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [ |
| { |
| "start": 0, |
| "end": 8, |
| "text": "EQUATION", |
| "ref_id": "EQREF", |
| "raw_str": "\u03b4 k = (t k \u2212 y k ) \u00d7 y k \u00d7 (1 \u2212 y k )", |
| "eq_num": "(5)" |
| } |
| ], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "EQUATION", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [ |
| { |
| "start": 0, |
| "end": 8, |
| "text": "EQUATION", |
| "ref_id": "EQREF", |
| "raw_str": "\u03b4 j = ( K \u2211 k=1 w jk \u00d7 \u03b4 k ) \u00d7 y k \u00d7 (1 \u2212 y k )", |
| "eq_num": "(6)" |
| } |
| ], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "\u8abf\u6574\u96b1\u85cf\u5c64\u5230\u8f38\u51fa\u5c64(7)\u53ca\u8f38\u5165\u5c64\u5230\u96b1\u85cf\u5c64(8)\u7684\u6b0a\u91cd\uff0cx i :\u8f38\u5165\u5c64\u7684\u795e\u7d93\u5143\uff0cz j :\u96b1\u85cf\u5c64 \u7684\u795e\u7d93\u5143\uff0cy k :\u8f38\u51fa\u5c64\u7684\u795e\u7d93\u5143 \u03b7\uff1alearning rate , m\uff1amomentum press, 1997.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "EQUATION", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [ |
| { |
| "start": 0, |
| "end": 8, |
| "text": "EQUATION", |
| "ref_id": "EQREF", |
| "raw_str": "w jk = w jk + \u03b7 \u00d7 \u03b4 k \u00d7 z j + m \u00d7 \u2206w previous (7) w i j = w i j + \u03b7 \u00d7 \u03b4 j \u00d7 x i + m \u00d7 \u2206w previous", |
| "eq_num": "(8" |
| } |
| ], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "[3] S. Steidl, \"Automatic classification of emotion related user states in spontaneous children'sspeech,\" PhD thesis, University of Erlangen-Nuremberg, 2009.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| } |
| ], |
| "back_matter": [], |
| "bib_entries": { |
| "BIBREF0": { |
| "ref_id": "b0", |
| "title": "A database of german emotional speech", |
| "authors": [ |
| { |
| "first": "F", |
| "middle": [], |
| "last": "Burkhardt", |
| "suffix": "" |
| }, |
| { |
| "first": "A", |
| "middle": [], |
| "last": "Paeschke", |
| "suffix": "" |
| }, |
| { |
| "first": "M", |
| "middle": [], |
| "last": "Rolfes", |
| "suffix": "" |
| }, |
| { |
| "first": "W", |
| "middle": [ |
| "F" |
| ], |
| "last": "Sendlmeier", |
| "suffix": "" |
| }, |
| { |
| "first": "B", |
| "middle": [], |
| "last": "Weiss", |
| "suffix": "" |
| } |
| ], |
| "year": 2005, |
| "venue": "in Interspeech", |
| "volume": "5", |
| "issue": "", |
| "pages": "1517--1520", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "F. Burkhardt, A. Paeschke, M. Rolfes, W. F. Sendlmeier, and B. Weiss, \"A database of german emotional speech.\" in Interspeech, vol. 5, 2005, pp. 1517-1520.", |
| "links": null |
| }, |
| "BIBREF1": { |
| "ref_id": "b1", |
| "title": "Speech Emotion Recognition via Nonlinear Dynamical Features)", |
| "authors": [ |
| { |
| "first": "Chen", |
| "middle": [], |
| "last": "Lin Chu-Hsuan", |
| "suffix": "" |
| }, |
| { |
| "first": "Yen", |
| "middle": [ |
| "-" |
| ], |
| "last": "Sheng", |
| "suffix": "" |
| }, |
| { |
| "first": "\"", |
| "middle": [], |
| "last": "\u7d50 \u5408 \u975e \u7dda \u6027 \u52d5 \u614b \u7279 \u5fb5 \u4e4b \u8a9e \u97f3 \u60c5 \u7dd2 \u8fa8 \u8b58", |
| "suffix": "" |
| } |
| ], |
| "year": 2015, |
| "venue": "", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "LIN Chu-Hsuan, CHEN, Yen-Sheng, \"\u7d50 \u5408 \u975e \u7dda \u6027 \u52d5 \u614b \u7279 \u5fb5 \u4e4b \u8a9e \u97f3 \u60c5 \u7dd2 \u8fa8 \u8b58 (Speech Emotion Recognition via Nonlinear Dynamical Features)\"[In Chinese], in RO- CLING 2015.", |
| "links": null |
| }, |
| "BIBREF2": { |
| "ref_id": "b2", |
| "title": "GMM supervector basedSVM with spectral features for speech emotion recognition", |
| "authors": [ |
| { |
| "first": "", |
| "middle": [], |
| "last": "Hu", |
| "suffix": "" |
| }, |
| { |
| "first": "Ming-Xing", |
| "middle": [], |
| "last": "Hao", |
| "suffix": "" |
| }, |
| { |
| "first": "Wei", |
| "middle": [], |
| "last": "Xu", |
| "suffix": "" |
| }, |
| { |
| "first": "", |
| "middle": [], |
| "last": "Wu", |
| "suffix": "" |
| } |
| ], |
| "year": 2007, |
| "venue": "Acoustics, Speech and Signal Processing", |
| "volume": "4", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Hu, Hao, Ming-Xing Xu, and Wei Wu. \"GMM supervector basedSVM with spectral fea- tures for speech emotion recognition.\"Acoustics, Speech and Signal Processing, 2007. ICASSP 2007.IEEE International Conference on. Vol. 4. IEEE, 2007.", |
| "links": null |
| }, |
| "BIBREF3": { |
| "ref_id": "b3", |
| "title": "Emulating humancognitive approach for speech emotion using MLP and Gen-SofNN", |
| "authors": [ |
| { |
| "first": "Norhaslinda", |
| "middle": [], |
| "last": "Kamaruddin", |
| "suffix": "" |
| }, |
| { |
| "first": "Abdul", |
| "middle": [], |
| "last": "Wahab", |
| "suffix": "" |
| } |
| ], |
| "year": 2013, |
| "venue": "Information and Communication Technology for theMuslim World (ICT4M), 2013 5th International Conference on", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Kamaruddin, Norhaslinda, and Abdul Wahab. \"Emulating humancognitive approach for speech emotion using MLP and Gen-SofNN.\" Information and Communication Technol- ogy for theMuslim World (ICT4M), 2013 5th International Conference on.IEEE, 2013.", |
| "links": null |
| }, |
| "BIBREF4": { |
| "ref_id": "b4", |
| "title": "Speech emotion recognition using gaussian mixture model", |
| "authors": [ |
| { |
| "first": "Xianglin", |
| "middle": [], |
| "last": "Cheng", |
| "suffix": "" |
| }, |
| { |
| "first": "Qiong", |
| "middle": [], |
| "last": "Duan", |
| "suffix": "" |
| } |
| ], |
| "year": 2012, |
| "venue": "The 2nd International Conference on Computer Application and System Modeling", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Cheng, Xianglin, and Qiong Duan. \"Speech emotion recognition using gaussian mixture model.\" The 2nd International Conference on Computer Application and System Model- ing. 2012. APA", |
| "links": null |
| }, |
| "BIBREF5": { |
| "ref_id": "b5", |
| "title": "A hierarchical framework for modeling multimodality and emotional evolution in affective dialogs", |
| "authors": [ |
| { |
| "first": "Angeliki", |
| "middle": [], |
| "last": "Metallinou", |
| "suffix": "" |
| }, |
| { |
| "first": "Athanasios", |
| "middle": [], |
| "last": "Katsamanis", |
| "suffix": "" |
| }, |
| { |
| "first": "Shrikanth", |
| "middle": [], |
| "last": "Narayanan", |
| "suffix": "" |
| } |
| ], |
| "year": null, |
| "venue": "", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Metallinou, Angeliki, Athanasios Katsamanis, and Shrikanth Narayanan. \"A hierarchi- cal framework for modeling multimodality and emotional evolution in affective dialogs.\"", |
| "links": null |
| }, |
| "BIBREF6": { |
| "ref_id": "b6", |
| "title": "2012 IEEE International Conference on", |
| "authors": [], |
| "year": 2012, |
| "venue": "", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Acoustics, Speech and Signal Processing (ICASSP), 2012 IEEE International Conference on. IEEE, 2012.", |
| "links": null |
| }, |
| "BIBREF7": { |
| "ref_id": "b7", |
| "title": "Neural networks and physical systems with emergent collective computational abilities", |
| "authors": [ |
| { |
| "first": "John", |
| "middle": [ |
| "J" |
| ], |
| "last": "Hopfield", |
| "suffix": "" |
| } |
| ], |
| "year": 1982, |
| "venue": "Proceedings of the national academy of sciences 79", |
| "volume": "8", |
| "issue": "", |
| "pages": "2554--2558", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Hopfield, John J. \"Neural networks and physical systems with emergent collective com- putational abilities.\" Proceedings of the national academy of sciences 79.8 (1982): 2554- 2558.", |
| "links": null |
| }, |
| "BIBREF8": { |
| "ref_id": "b8", |
| "title": "The INTERSPEECH 2009 emotion challenge", |
| "authors": [ |
| { |
| "first": "Bj\u00f6rn", |
| "middle": [ |
| "W" |
| ], |
| "last": "Schuller", |
| "suffix": "" |
| }, |
| { |
| "first": "Stefan", |
| "middle": [], |
| "last": "Steidl", |
| "suffix": "" |
| }, |
| { |
| "first": "Anton", |
| "middle": [], |
| "last": "Batliner", |
| "suffix": "" |
| } |
| ], |
| "year": 2009, |
| "venue": "", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Schuller, Bj\u00f6rn W., Stefan Steidl, and Anton Batliner. \"The INTERSPEECH 2009 emo- tion challenge.\" Interspeech. Vol. 2009. 2009", |
| "links": null |
| }, |
| "BIBREF9": { |
| "ref_id": "b9", |
| "title": "Cross-lingual automatic speech emotion recognition", |
| "authors": [ |
| { |
| "first": "B.-C", |
| "middle": [], |
| "last": "Chiou", |
| "suffix": "" |
| } |
| ], |
| "year": 2014, |
| "venue": "", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "B.-C. Chiou, \"Cross-lingual automatic speech emotion recognition,\"Master's thesis, Na- tional Sun Yat-sen University, 2014", |
| "links": null |
| }, |
| "BIBREF10": { |
| "ref_id": "b10", |
| "title": "Skewness-Robust Neural Networks with Application to Speech Emotion Recognition", |
| "authors": [ |
| { |
| "first": "P.-Y", |
| "middle": [], |
| "last": "Shih", |
| "suffix": "" |
| } |
| ], |
| "year": 2016, |
| "venue": "", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "P.-Y.Shih, Skewness-Robust Neural Networks with Application to Speech Emotion Recognition, Master's thesis, National Sun Yat-sen University, 2016.", |
| "links": null |
| }, |
| "BIBREF11": { |
| "ref_id": "b11", |
| "title": "Recognising realistic emotions and affect in speech: State of the art and lessons learnt from the first challenge", |
| "authors": [ |
| { |
| "first": "", |
| "middle": [], |
| "last": "Schuller", |
| "suffix": "" |
| }, |
| { |
| "first": "", |
| "middle": [], |
| "last": "Bj\u00f6rn", |
| "suffix": "" |
| } |
| ], |
| "year": 2011, |
| "venue": "Speech Communication", |
| "volume": "53", |
| "issue": "", |
| "pages": "1062--1087", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Schuller, Bj\u00f6rn, et al. \"Recognising realistic emotions and affect in speech: State of the art and lessons learnt from the first challenge.\" Speech Communication 53.9 (2011): 1062- 1087.", |
| "links": null |
| }, |
| "BIBREF12": { |
| "ref_id": "b12", |
| "title": "Brno university of technology system for interspeech 2009 emotion challenge", |
| "authors": [ |
| { |
| "first": "Marcel", |
| "middle": [], |
| "last": "Kockmann", |
| "suffix": "" |
| }, |
| { |
| "first": "Luk\u00e1\u0161", |
| "middle": [], |
| "last": "Burget", |
| "suffix": "" |
| }, |
| { |
| "first": "Jan\u010dernock\u00fd", |
| "middle": [], |
| "last": "", |
| "suffix": "" |
| } |
| ], |
| "year": 2009, |
| "venue": "Tenth Annual Conference of the International Speech Communication Association", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Kockmann, Marcel, Luk\u00e1\u0161 Burget, and Jan\u010cernock\u00fd. \"Brno university of technology system for interspeech 2009 emotion challenge.\" Tenth Annual Conference of the Inter- national Speech Communication Association. 2009.", |
| "links": null |
| } |
| }, |
| "ref_entries": { |
| "TABREF1": { |
| "html": null, |
| "num": null, |
| "text": "Van Bezooijen, Ren\u00e9e, Stanley A. Otto, and Thomas A. Heenan. \"Recognition of vocal expressions of emotion: A three-nation study to identify universal characteristics.\" Journal of Cross-Cultural Psychology 14.4 (1983): 387-406.[2] Picard, Rosalind W., and Roalind Picard. Affective computing. Vol. 252. Cambridge: MIT", |
| "content": "<table><tr><td colspan=\"7\">\u8868 \u4e8c\u3001 FAU-Aibo\u60c5\u7dd2\u8a9e\u6599\u5eab \u8868 \u4e94\u3001 Skewness-robust MLP \u5206\u985e\u7d50\u679c\u6df7\u6dc6\u77e9\u9663</td></tr><tr><td/><td/><td colspan=\"5\">Angry Emphatic Neutral Positive Rest Angry Emphatic Neutral Positive Rest Recall</td></tr><tr><td colspan=\"7\">Train Angry 674 Test 881 2093 5590 300 131 70 30 611 1508 5377 215 Emphatic 218 778 281 52</td><td>80 546 179 51.6% 49.1%</td></tr><tr><td>Neutral</td><td/><td>528</td><td colspan=\"2\">900</td><td>2209</td><td>666</td><td>1074 41.1%</td></tr><tr><td colspan=\"7\">\u80fd\u91cf(RMS energy)\u3001\u904e\u96f6\u7387(Zero Crossing Rate, ZCR)\u3001\u8ae7\u97f3\u566a\u97f3\u6bd4(Harmonics-to-Noise Positive 11 10 29 116 49 54.0%</td></tr><tr><td colspan=\"7\">Ratio, HNR)\u3001\u97f3\u9ad8\u983b\u7387(Pitch Frequency)\uff0c\u52a0\u4e0a\u6bcf\u500b\u4f4e\u968e\u53c3\u6578\u7684\u4e00\u968e\u4fc2\u6578\u5dee(Delta)\u300212 Rest 300 79 121 104 150 27.5%</td></tr><tr><td colspan=\"7\">\u614b(Skewness)\u3001\u6700\u5927\u6700\u5c0f\u503c\u3001\u76f8\u5c0d\u4f4d\u7f6e(Relative Position)\u3001\u7bc4\u570d(Range)\u4ee5\u53ca\u53e6\u5916\u5169\u500b\u7dda \u5716 \u4e09\u3001 MLP\u67b6\u69cb\u5716 \u500b \u6cdb \u51fd(Functionals)\u70ba \uff1a \u5e73 \u5747 \u503c \u3001 \u6a19 \u6e96 \u5dee(standard deviation)\u3001 \u5cf0 \u5ea6(Kurtosis)\u548c \u504f \u79fb Avg.recall 44.6%</td></tr><tr><td colspan=\"7\">\u5716 \u4e8c\u3001 \u8a9e\u8005\u6b63\u898f\u5316\u6d41\u7a0b \u6027\u8ff4\u6b78\u4fc2\u6578(Linear Regression Coefficients)\u53ca\u5176\u5747\u65b9\u5dee(Mean Square Error, MSE)\u3002\u56e0\u6b64\uff0c \u8868 \u4e09\u3001 \u5be6\u9a57\u53c3\u6578 \u8868 \u516d\u3001 Teacher-student model \u5206\u985e\u7d50\u679c\u6df7\u6dc6\u77e9\u9663</td></tr><tr><td colspan=\"7\">\u8868 \u4e00\u3001 \u985e\u5225\u6b0a\u91cd \u5c0d\u65bc\u6bcf\u4e00\u500b\u4f4e\u968e\u53c3\u6578\uff0c\u7d93\u904e\u4e00\u968e\u4fc2\u6578\u5dee\u8a08\u7b97\u518d\u7d93\u753112\u500b\u6cdb\u51fd\u8a08\u7b97\u5f8c\uff0c\u6700\u5f8c\u5f97\u5230\u7684\u7279\u5fb5</td></tr><tr><td colspan=\"7\">\u96c6\u5305\u542b\u4e86 16\u00d72\u00d712=384 \u7dad\u7279\u5fb5\u53c3\u6578\u3002 Hyperparameter Angry Emphatic Neutral Positive Rest Recall Value</td></tr><tr><td colspan=\"7\">Angry Emphatic Neutral Positive Rest \u5728\u8a13\u7df4\u7db2\u8def\u4e4b\u524d\uff0c\u6703\u5148\u5c07\u8a13\u7df4\u8cc7\u6599\u6b63\u898f\u5316\u5230 [0,1] \u4ee5\u964d\u4f4e\u539f\u59cb\u8cc7\u6599\u9593\u7684\u5dee\u7570\u6027\u3002 Mini-batch 100 Angry 329 110 72 37 63 53.8%</td></tr><tr><td colspan=\"5\">Weight Learning rate 1.1 Emphatic 265 776 0.5</td><td colspan=\"2\">0.2 278</td><td>0.4 92</td><td>1.5</td><td>1.4 97 51.5%</td></tr><tr><td>Neutral</td><td colspan=\"4\">Learning rate decay 630 948</td><td>2085</td><td>0.0005 1073</td><td>641 38.8%</td></tr><tr><td colspan=\"2\">Positive</td><td colspan=\"2\">Momentum 8</td><td>7</td><td>32</td><td>0.5 141</td><td>27 65.6%</td></tr><tr><td>\u4e09\u3001 \u5be6\u9a57\u7d50\u679c Rest</td><td/><td colspan=\"2\">Optimizer 80</td><td>86</td><td colspan=\"2\">Stochastic gradient descent 117 151 112 20.5%</td></tr><tr><td/><td/><td colspan=\"3\">Loss function</td><td colspan=\"2\">Cross-entropy Avg.recall 46.0%</td></tr><tr><td colspan=\"7\">\u7387\uff0c\u5be6\u9a57\u7d50\u679c\u5982\u8868\u56db\u3002\u5404\u985e\u60c5\u7dd2\u5206\u985e\u60c5\u5f62\u5982\u8868\u4e94\u3001\u8868\u516d\u3002\u672c\u7814\u7a76\u4f7f\u7528 teacher-student Epoch 600 ) \u8cc7\u6599\u524d\u8655\u7406\u7684\u90e8\u4efd\uff0c\u7531\u65bc\u4e0d\u540c\u8a9e\u8005\u7522\u751f\u7684\u8072\u97f3\u6703\u6709\u6240\u5dee\u7570\uff0c\u56e0\u6b64\u672c\u5be6\u9a57\u6703\u4f7f\u7528\u8a9e model \u6240\u5f97\u5230\u7684\u8fa8\u8b58\u7387(46%)\u6bd4\u57fa\u6e96\u8fa8\u8b58\u7387(38.2%)\u9ad8\u51fa\u7d048%\uff0c\u6b64\u5916\uff0c\u6839\u64da Interspeech \u53c3\u8003\u6587\u737b 2009 Emotion Challenge \u53c3\u8cfd\u8005\u6240\u5f97\u5230\u7684\u591a\u7d44\u5be6\u9a57\u7d50\u679c[14]\u4e2d\uff0c\u6700\u4f73\u7684\u7d50\u679c\u70baMarcel \u8868 \u56db\u3001 MLP\u3001Teacher-Student Model \u5be6\u9a57\u7d50\u679c \u8005\u6b63\u898f\u5316\u7684\u65b9\u6cd5\u4f86\u6d88\u9664\u6b64\u5dee\u7570\u6027\uff0c\u4e26\u53ea\u4fdd\u7559\u60c5\u7dd2\u7684\u8b8a\u7570\u3002\u8a9e\u8005\u6b63\u898f\u5316\u6703\u5c07\u591a\u500b\u5be6\u969b\u8a9e \u8005\u8f49\u63db\u70ba\u4e00\u500b\u865b\u64ec\u8a9e\u8005\uff0c\u5982\u6b64\u4e00\u4f86\u6211\u5011\u5c31\u80fd\u5920\u5f97\u5230\u4e00\u500b\u865b\u64ec\u8a9e\u8005\u7684\u8cc7\u6599\u5206\u5e03\uff0c\u63a5\u4e0b\u4f86 \u6821\u3002\u8a9e\u6599\u5eab\u7684\u60c5\u7dd2\u6a19\u8a18\u5de5\u4f5c\u7531 5 \u540d\u5c08\u696d\u7684\u8a9e\u8a00\u5b78\u8005\u5171\u540c\u5b8c\u6210\uff0c\u5171\u5206\u70ba11\u985e\u60c5\u7dd2\uff0c\u5206\u5225 Kockmann \u7b49\u4eba[15]\u6240\u7372\u5f97\u7684\u768441.65%\u3002 [1]</td></tr><tr><td colspan=\"7\">\u5c07\u6bcf\u500b\u5be6\u969b\u8a9e\u8005\u90fd\u8f49\u63db\u6210\u865b\u64ec\u8a9e\u8005\u7684\u5206\u5e03\uff0c\u6b63\u898f\u5316\u7684\u65b9\u6cd5\u70ba\u76f4\u65b9\u5716\u5747\u8861\u6cd5(Histogram \u70ba\uff1a\u6b61\u6a02(Joyful)\u3001\u9a5a\u8a1d(Surprised)\u3001\u5f37\u8abf(Emphatic)\u3001\u7121\u5948(Helpless)\u3001\u654f\u611f(Touchy)\u3001 \u56db\u3001 \u7d50\u8ad6 Recall</td></tr><tr><td colspan=\"7\">Equalization, HE)[12]\u3002\u8a9e\u8005\u6b63\u898f\u5316\u7684\u6d41\u7a0b\u5982\u5716\u4e8c\u3002 \u61a4\u6012(Angry)\u3001\u5abd\u5abd\u8a9e(Motherese)\u3001\u7121\u804a(Bored)\u3001\u8b74\u8cac(Reprimanding)\u3001\u4e2d\u6027(Neutral)\u8207 Skewness-robust MLP 44.6%</td></tr><tr><td colspan=\"7\">\u70ba\u4e86\u8655\u7406\u5404\u985e\u8a13\u7df4\u8cc7\u6599\u4e0d\u5e73\u8861\u7684\u554f\u984c\uff0c\u5c0d\u65bc\u6bcf\u4e00\u985e\u5225\uff0c\u672c\u7814\u7a76\u53c3\u8003[13]\u5f15\u5165\u4e00\u500b\u985e \u6b63\u5411 (Positive)\u3002\u672c\u5be6\u9a57\u4f9d\u7167 Interspeech 2009 Emotion Challenge\u7684\u60c5\u611f\u8b58\u5225\u6311\u6230\uff0c\u9078\u51fa \u6839\u64da\u8868\u56db\u6240\u5f97\u5230\u7684\u5be6\u9a57\u7d50\u679c\u986f\u793a\uff0c\u4f7f\u7528 teacher-student model\u4e4b\u5f8c\uff0c\u8fa8\u8b58\u7387\u80fd\u5920 Teacher-student model 46.0%</td></tr><tr><td colspan=\"7\">\u61a4\u6012(Angry)\u3001\u5f37\u8abf(Emphatic)\u3001\u4e2d\u6027(Neutral)\u3001\u6b63\u9762(Positive)\u3001\u5176\u9918(Rest)\u4e94\u985e\u60c5\u7dd2\u3002\u6b64 \u5f9e44.6%\u63d0\u5347\u523046%\u3002\u56e0\u6b64\u53ef\u5f97\u77e5\u539f\u672c\u7528\u4eba\u70ba\u6a19\u8a18\u7684\u65b9\u5f0f\u5b58\u5728\u4e00\u4e9b\u554f\u984c\uff0c\u53ef\u80fd\u6703\u9020\u6210 \u5225\u6b0a\u91cd r ik \u4f86\u8abf\u6574\u53c3\u6578\u66f4\u65b0\u3002\u5176\u4e2d\uff0cr ik \u70ba\u8a72\u985e\u5225\u4e4b\u8a13\u7df4\u8cc7\u6599\u6578\u8207\u6574\u9ad4\u8cc7\u6599\u6578\u4e4b\u76f8\u5c0d\u983b\u7387 \u7684\u5012\u6578\uff0c\u8207\u8a72\u7b46\u8cc7\u6599\u6240\u5c6c\u985e\u5225\u4e4b\u7e3d\u8cc7\u6599\u6578\u6210\u53cd\u6bd4(9)\u3002\u52a0\u5165\u985e\u5225\u6b0a\u91cd\u9032\u884c\u8a13\u7df4\u7684MLP\u6a21 \u578b\u5373\u70baSkewness-robust MLP\u3002 r ik = N N k \u221d 1 N k (9) MLP \u5728\u5b78\u7fd2\u6642\uff0c\u7121\u6cd5\u91dd\u5c0d\u8cc7\u6599\u7684\u7279\u5fb5\u503c\u9032\u884c\u5b78\u7fd2\uff0c\u800c\u5728\u4f7f\u7528 teacher label \u6539\u8b8a\u539f\u672c\u7684 \u8cc7\u6599\u505a\u6a19\u8a18\u6216\u4fee\u6539\uff0c\u53ef\u80fd\u6703\u9047\u5230\u7684\u6311\u6230\u5305\u542b\u6a19\u8a18\u7684\u7d50\u679c\u662f\u5426\u5177\u6709\u8db3\u5920\u7684\u53ef\u9760\u6027\uff0c\u4ee5\u53ca \u6a19\u7c64\u5f8c\uff0c\u6709\u52a9\u65bc\u63d0\u5347 MLP \u5c0d FAU-Aibo \u60c5\u7dd2\u8a9e\u6599\u5eab\u7684\u8fa8\u8b58\u7387\u3002\u6709\u9451\u65bc\u5c0d\u8cc7\u6599\u7684\u6a19\u7c64\u505a \u4f7f\u7528\u6b64\u6a19\u7c64\u9032\u884c\u8a13\u7df4\u6642\uff0c\u5c0d\u65bc\u795e\u7d93\u7db2\u8def\u8a13\u7df4\u904e\u7a0b\u7684\u5f71\u97ff\u3002\u5e0c\u671b\u80fd\u5920\u7d50\u5408\u76f8\u95dc\u7684\u8cc7\u6599\u6a19 \u4fee\u6539\u80fd\u5920\u63d0\u5347\u8fa8\u8b58\u7387\uff0c\u56e0\u6b64\u5728\u672a\u4f86\u7684\u7814\u7a76\u4e2d\uff0c\u6211\u5011\u60f3\u9032\u4e00\u6b65\u7684\u53bb\u7814\u7a76\u8cc7\u6599\u6a19\u7c64\u7684\u6a19\u8a18 \u65b9\u6cd5\uff0c\u82e5\u80fd\u4ee5\u5176\u4ed6\u65b9\u5f0f\u7d50\u5408 teacher-student training \uff0c\u6216\u8a31\u80fd\u5728\u66f4\u77ed\u7684\u6642\u9593\u5167\u5c0d\u5927\u91cf\u7684 \u8a18\u65b9\u6cd5\u4f86\u8b93 MLP \u5c0d\u65bc\u4e94\u985e FAU-Aibo \u60c5\u7dd2\u8a9e\u6599\u5eab\u7684\u8fa8\u8b58\u7387\u63d0\u5347\u3002</td></tr><tr><td colspan=\"3\">\u5404\u985e\u5225\u7684\u6b0a\u91cd\u5982\u8868\u4e00\u6240\u793a\u3002</td><td/><td/><td/></tr></table>", |
| "type_str": "table" |
| } |
| } |
| } |
| } |
