| { | |
| "paper_id": "2019", | |
| "header": { | |
| "generated_with": "S2ORC 1.0.0", | |
| "date_generated": "2023-01-19T14:54:43.896693Z" | |
| }, | |
| "title": "A Preliminary Study on Deep Learning Neural Networks-based Multi-Model Sentiment Detection", | |
| "authors": [ | |
| { | |
| "first": "Tai-Rong", | |
| "middle": [], | |
| "last": "\u9673\u6cf0\u878d", | |
| "suffix": "", | |
| "affiliation": { | |
| "laboratory": "", | |
| "institution": "National Taipei University of Technology", | |
| "location": {} | |
| }, | |
| "email": "" | |
| }, | |
| { | |
| "first": "Yuan-Fu", | |
| "middle": [], | |
| "last": "Chen", | |
| "suffix": "", | |
| "affiliation": { | |
| "laboratory": "", | |
| "institution": "National Taipei University of Technology", | |
| "location": {} | |
| }, | |
| "email": "" | |
| }, | |
| { | |
| "first": "\u570b\u7acb\u81fa\u5317\u79d1\u6280\u5927\u5b78\u96fb\u5b50\u5de5\u7a0b\u7cfb", | |
| "middle": [], | |
| "last": "Liao", | |
| "suffix": "", | |
| "affiliation": { | |
| "laboratory": "", | |
| "institution": "National Taipei University of Technology", | |
| "location": {} | |
| }, | |
| "email": "[email protected]" | |
| }, | |
| { | |
| "first": "Chen-Ming", | |
| "middle": [], | |
| "last": "\u6f58\u632f\u9298", | |
| "suffix": "", | |
| "affiliation": { | |
| "laboratory": "", | |
| "institution": "Telecommunication Laboratories", | |
| "location": { | |
| "settlement": "Chunghwa Telecom", | |
| "country": "Taoyuan Taiwan" | |
| } | |
| }, | |
| "email": "[email protected]" | |
| }, | |
| { | |
| "first": "Tzu-Hsiu", | |
| "middle": [], | |
| "last": "Pan", | |
| "suffix": "", | |
| "affiliation": { | |
| "laboratory": "", | |
| "institution": "Telecommunication Laboratories", | |
| "location": { | |
| "settlement": "Chunghwa Telecom", | |
| "country": "Taoyuan Taiwan" | |
| } | |
| }, | |
| "email": "" | |
| }, | |
| { | |
| "first": "", | |
| "middle": [], | |
| "last": "Kuo", | |
| "suffix": "", | |
| "affiliation": { | |
| "laboratory": "", | |
| "institution": "Telecommunication Laboratories", | |
| "location": { | |
| "settlement": "Chunghwa Telecom", | |
| "country": "Taoyuan Taiwan" | |
| } | |
| }, | |
| "email": "" | |
| }, | |
| { | |
| "first": "Matu\u0301s\u030c", | |
| "middle": [], | |
| "last": "Pleva", | |
| "suffix": "", | |
| "affiliation": { | |
| "laboratory": "", | |
| "institution": "Technical University of Kos\u030cice", | |
| "location": { | |
| "country": "Slovakia" | |
| } | |
| }, | |
| "email": "[email protected]" | |
| }, | |
| { | |
| "first": "Daniel", | |
| "middle": [], | |
| "last": "Hla\u0301dek", | |
| "suffix": "", | |
| "affiliation": { | |
| "laboratory": "", | |
| "institution": "Technical University of Kos\u030cice", | |
| "location": { | |
| "country": "Slovakia" | |
| } | |
| }, | |
| "email": "[email protected]" | |
| } | |
| ], | |
| "year": "", | |
| "venue": null, | |
| "identifiers": {}, | |
| "abstract": "", | |
| "pdf_parse": { | |
| "paper_id": "2019", | |
| "_pdf_hash": "", | |
| "abstract": [], | |
| "body_text": [], | |
| "back_matter": [], | |
| "bib_entries": { | |
| "BIBREF0": { | |
| "ref_id": "b0", | |
| "title": "An investigation of speech-based human emotion recognition", | |
| "authors": [ | |
| { | |
| "first": "Y", | |
| "middle": [], | |
| "last": "Wang", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "L", | |
| "middle": [], | |
| "last": "Guan", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2004, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "15--18", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Y. Wang, L. Guan, An investigation of speech-based human emotion recognition, pp. 15- 18, 2004.", | |
| "links": null | |
| }, | |
| "BIBREF2": { | |
| "ref_id": "b2", | |
| "title": "Recognizing human emotional state from audiovisual signals", | |
| "authors": [ | |
| { | |
| "first": "Y", | |
| "middle": [], | |
| "last": "Wang", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "L", | |
| "middle": [], | |
| "last": "Guan", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2008, | |
| "venue": "IEEE Trans. Multimedia", | |
| "volume": "10", | |
| "issue": "5", | |
| "pages": "936--946", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Y. Wang, L. Guan, \"Recognizing human emotional state from audiovisual signals\", IEEE Trans. Multimedia, vol. 10, no. 5, pp. 936-946, Aug. 2008.", | |
| "links": null | |
| }, | |
| "BIBREF3": { | |
| "ref_id": "b3", | |
| "title": "Audio-visual affective expression recognition through multistream fused HMM", | |
| "authors": [ | |
| { | |
| "first": "Z", | |
| "middle": [], | |
| "last": "Zeng", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "J", | |
| "middle": [], | |
| "last": "Tu", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "B", | |
| "middle": [ | |
| "M" | |
| ], | |
| "last": "Pianfetti", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "T", | |
| "middle": [ | |
| "S" | |
| ], | |
| "last": "Huang", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2008, | |
| "venue": "IEEE Trans. Multimedia", | |
| "volume": "10", | |
| "issue": "4", | |
| "pages": "570--577", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Z. Zeng, J. Tu, B. M. Pianfetti, T. S. Huang, \"Audio-visual affective expression recognition through multistream fused HMM\", IEEE Trans. Multimedia, vol. 10, no. 4, pp. 570-577, Jun. 2008.", | |
| "links": null | |
| }, | |
| "BIBREF4": { | |
| "ref_id": "b4", | |
| "title": "Multimodal information fusion application to human emotion recognition from face and speech", | |
| "authors": [ | |
| { | |
| "first": "M", | |
| "middle": [], | |
| "last": "Mansoorizadeh", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "N", | |
| "middle": [ | |
| "M" | |
| ], | |
| "last": "Charkari", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2010, | |
| "venue": "Multimedia Tools Appl", | |
| "volume": "49", | |
| "issue": "2", | |
| "pages": "277--297", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "M. Mansoorizadeh, N. M. Charkari, \"Multimodal information fusion application to human emotion recognition from face and speech\", Multimedia Tools Appl., vol. 49, no. 2, pp. 277-297, 2010.", | |
| "links": null | |
| }, | |
| "BIBREF5": { | |
| "ref_id": "b5", | |
| "title": "Multiple classifier systems for the classification of audio-visual emotional states", | |
| "authors": [ | |
| { | |
| "first": "M", | |
| "middle": [], | |
| "last": "Glodek", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2011, | |
| "venue": "Affective Computing and Intelligent Interaction", | |
| "volume": "6975", | |
| "issue": "", | |
| "pages": "359--368", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "M. Glodek et al., \"Multiple classifier systems for the classification of audio-visual emotional states\" in Affective Computing and Intelligent Interaction, Berlin, Germany:Springer, vol. 6975, pp. 359-368, 2011.", | |
| "links": null | |
| }, | |
| "BIBREF6": { | |
| "ref_id": "b6", | |
| "title": "Multimodal emotion recognition in response to videos", | |
| "authors": [ | |
| { | |
| "first": "M", | |
| "middle": [], | |
| "last": "Soleymani", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "M", | |
| "middle": [], | |
| "last": "Pantic", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "T", | |
| "middle": [], | |
| "last": "Pun", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2012, | |
| "venue": "IEEE Trans. Affect. Comput", | |
| "volume": "3", | |
| "issue": "2", | |
| "pages": "211--223", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "M. Soleymani, M. Pantic, T. Pun, \"Multimodal emotion recognition in response to videos\", IEEE Trans. Affect. Comput., vol. 3, no. 2, pp. 211-223, Apr./Jun. 2012.", | |
| "links": null | |
| }, | |
| "BIBREF7": { | |
| "ref_id": "b7", | |
| "title": "Error weighted semi-coupled hidden Markov model for audio-visual emotion recognition", | |
| "authors": [ | |
| { | |
| "first": "J.-C", | |
| "middle": [], | |
| "last": "Lin", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "C.-H", | |
| "middle": [], | |
| "last": "Wu", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "W.-L", | |
| "middle": [], | |
| "last": "Wei", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2012, | |
| "venue": "IEEE Trans. Multimedia", | |
| "volume": "14", | |
| "issue": "1", | |
| "pages": "142--156", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "J.-C. Lin, C.-H. Wu, W.-L. Wei, \"Error weighted semi-coupled hidden Markov model for audio-visual emotion recognition\", IEEE Trans. Multimedia, vol. 14, no. 1, pp. 142-156, Feb. 2012.", | |
| "links": null | |
| }, | |
| "BIBREF8": { | |
| "ref_id": "b8", | |
| "title": "Exploring fusion methods for multimodal emotion recognition with missing data", | |
| "authors": [ | |
| { | |
| "first": "J", | |
| "middle": [], | |
| "last": "Wagner", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "E", | |
| "middle": [], | |
| "last": "Andre", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "F", | |
| "middle": [], | |
| "last": "Lingenfelser", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "J", | |
| "middle": [], | |
| "last": "Kim", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2011, | |
| "venue": "IEEE Trans. Affect. Comput", | |
| "volume": "2", | |
| "issue": "4", | |
| "pages": "206--218", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "J. Wagner, E. Andre, F. Lingenfelser, J. Kim, \"Exploring fusion methods for multimodal emotion recognition with missing data\", IEEE Trans. Affect. Comput., vol. 2, no. 4, pp. 206-218, Oct. 2011.", | |
| "links": null | |
| }, | |
| "BIBREF9": { | |
| "ref_id": "b9", | |
| "title": "Context-sensitive learning for enhanced audiovisual emotion classification", | |
| "authors": [ | |
| { | |
| "first": "A", | |
| "middle": [], | |
| "last": "Metallinou", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "M", | |
| "middle": [], | |
| "last": "W\u00f6llmer", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "A", | |
| "middle": [], | |
| "last": "Katsamanis", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "F", | |
| "middle": [], | |
| "last": "Eyben", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "B", | |
| "middle": [], | |
| "last": "Schuller", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "S", | |
| "middle": [], | |
| "last": "Narayanan", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2012, | |
| "venue": "IEEE Trans. Affect. Comput", | |
| "volume": "3", | |
| "issue": "2", | |
| "pages": "184--198", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "A. Metallinou, M. W\u00f6llmer, A. Katsamanis, F. Eyben, B. Schuller, S. Narayanan, \"Context-sensitive learning for enhanced audiovisual emotion classification\", IEEE Trans. Affect. Comput., vol. 3, no. 2, pp. 184-198, Apr./Jun. 2012.", | |
| "links": null | |
| }, | |
| "BIBREF10": { | |
| "ref_id": "b10", | |
| "title": "Audio-visual emotion recognition using FCBF feature selection method and particle swarm optimization for fuzzy ARTMAP neural networks", | |
| "authors": [ | |
| { | |
| "first": "D", | |
| "middle": [], | |
| "last": "Gharavian", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "M", | |
| "middle": [], | |
| "last": "Bejani", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "M", | |
| "middle": [], | |
| "last": "Sheikhan", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2017, | |
| "venue": "Multimedia Tools Appl", | |
| "volume": "76", | |
| "issue": "2", | |
| "pages": "2331--2352", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "D. Gharavian, M. Bejani, M. Sheikhan, \"Audio-visual emotion recognition using FCBF feature selection method and particle swarm optimization for fuzzy ARTMAP neural networks\", Multimedia Tools Appl., vol. 76, no. 2, pp. 2331-2352, 2017.", | |
| "links": null | |
| }, | |
| "BIBREF11": { | |
| "ref_id": "b11", | |
| "title": "BAUM-1: A spontaneous audio-visual face database of affective and mental states", | |
| "authors": [ | |
| { | |
| "first": "S", | |
| "middle": [], | |
| "last": "Zhalehpour", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "O", | |
| "middle": [], | |
| "last": "Onder", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Z", | |
| "middle": [], | |
| "last": "Akhtar", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "C", | |
| "middle": [ | |
| "E" | |
| ], | |
| "last": "Erdem", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": null, | |
| "venue": "IEEE Trans. Affect. Comput", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "S. Zhalehpour, O. Onder, Z. Akhtar, C. E. Erdem, \"BAUM-1: A spontaneous audio-visual face database of affective and mental states\", IEEE Trans. Affect. Comput..", | |
| "links": null | |
| }, | |
| "BIBREF12": { | |
| "ref_id": "b12", | |
| "title": "FF-SKPCCA: Kernel probabilistic canonical correlation analysis", | |
| "authors": [ | |
| { | |
| "first": "R", | |
| "middle": [ | |
| "R" | |
| ], | |
| "last": "Sarvestani", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "R", | |
| "middle": [], | |
| "last": "Boostani", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2017, | |
| "venue": "Appl. Intell", | |
| "volume": "46", | |
| "issue": "2", | |
| "pages": "438--454", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "R. R. Sarvestani, R. Boostani, \"FF-SKPCCA: Kernel probabilistic canonical correlation analysis\", Appl. Intell., vol. 46, no. 2, pp. 438-454, 2017.", | |
| "links": null | |
| }, | |
| "BIBREF13": { | |
| "ref_id": "b13", | |
| "title": "Audiovisual emotion recognition using ANOVA feature selection method and multi-classifier neural networks", | |
| "authors": [ | |
| { | |
| "first": "M", | |
| "middle": [], | |
| "last": "Bejani", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "D", | |
| "middle": [], | |
| "last": "Gharavian", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "N", | |
| "middle": [ | |
| "M" | |
| ], | |
| "last": "Charkari", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2014, | |
| "venue": "Neural Comput. Appl", | |
| "volume": "24", | |
| "issue": "2", | |
| "pages": "399--412", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "M. Bejani, D. Gharavian, N. M. Charkari, \"Audiovisual emotion recognition using ANOVA feature selection method and multi-classifier neural networks\", Neural Comput. Appl., vol. 24, no. 2, pp. 399-412, 2014.", | |
| "links": null | |
| }, | |
| "BIBREF14": { | |
| "ref_id": "b14", | |
| "title": "Recognizing human emotional state from audiovisual signals", | |
| "authors": [ | |
| { | |
| "first": "Y", | |
| "middle": [], | |
| "last": "Wang", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "L", | |
| "middle": [], | |
| "last": "Guan", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2008, | |
| "venue": "IEEE Trans. Multimedia", | |
| "volume": "10", | |
| "issue": "5", | |
| "pages": "936--946", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Y. Wang, L. Guan, \"Recognizing human emotional state from audiovisual signals\", IEEE Trans. Multimedia, vol. 10, no. 5, pp. 936-946, Aug. 2008.", | |
| "links": null | |
| }, | |
| "BIBREF15": { | |
| "ref_id": "b15", | |
| "title": "Multimodal information fusion application to human emotion recognition from face and speech", | |
| "authors": [ | |
| { | |
| "first": "M", | |
| "middle": [], | |
| "last": "Mansoorizadeh", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "N", | |
| "middle": [ | |
| "M" | |
| ], | |
| "last": "Charkari", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2010, | |
| "venue": "Multimedia Tools Appl", | |
| "volume": "49", | |
| "issue": "2", | |
| "pages": "277--297", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "M. Mansoorizadeh, N. M. Charkari, \"Multimodal information fusion application to human emotion recognition from face and speech\", Multimedia Tools Appl., vol. 49, no. 2, pp. 277-297, 2010.", | |
| "links": null | |
| }, | |
| "BIBREF16": { | |
| "ref_id": "b16", | |
| "title": "Kernel cross-modal factor analysis for information fusion with application to bimodal emotion recognition", | |
| "authors": [ | |
| { | |
| "first": "Y", | |
| "middle": [], | |
| "last": "Wang", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "L", | |
| "middle": [], | |
| "last": "Guan", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "A", | |
| "middle": [ | |
| "N" | |
| ], | |
| "last": "Venetsanopoulos", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": null, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Y. Wang, L. Guan, A. N. Venetsanopoulos, \"Kernel cross-modal factor analysis for information fusion with application to bimodal emotion recognition\", IEEE Trans.", | |
| "links": null | |
| }, | |
| "BIBREF18": { | |
| "ref_id": "b18", | |
| "title": "Audiovisual recognition of spontaneous interest within conversations", | |
| "authors": [ | |
| { | |
| "first": "B", | |
| "middle": [], | |
| "last": "Schuller", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "R", | |
| "middle": [], | |
| "last": "M\u00fcller", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "B", | |
| "middle": [], | |
| "last": "H\u00f6rnler", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "A", | |
| "middle": [], | |
| "last": "H\u00f6thker", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "H", | |
| "middle": [], | |
| "last": "Konosu", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "G", | |
| "middle": [], | |
| "last": "", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2007, | |
| "venue": "Proc. 9th Int. Conf. Multimodal Interfaces (ICMI)", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "30--37", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "B. Schuller, R. M\u00fcller, B. H\u00f6rnler, A. H\u00f6thker, H. Konosu, G. Rigoll, \"Audiovisual recognition of spontaneous interest within conversations\", Proc. 9th Int. Conf. Multimodal Interfaces (ICMI), pp. 30-37, 2007.", | |
| "links": null | |
| }, | |
| "BIBREF19": { | |
| "ref_id": "b19", | |
| "title": "Analysis of emotion recognition using facial expressions speech and multimodal information", | |
| "authors": [ | |
| { | |
| "first": "C", | |
| "middle": [], | |
| "last": "Busso", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2004, | |
| "venue": "Proc. 6th Int. Conf. Multimodal Interfaces (ICMI)", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "205--211", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "C. Busso et al., \"Analysis of emotion recognition using facial expressions speech and multimodal information\", Proc. 6th Int. Conf. Multimodal Interfaces (ICMI), pp. 205-211, 2004.", | |
| "links": null | |
| } | |
| }, | |
| "ref_entries": { | |
| "TABREF0": { | |
| "num": null, | |
| "type_str": "table", | |
| "text": "\u95dc\u9375\u8a5e\uff1a\u60c5\u611f\u6aa2\u6e2c\uff0cCNN\uff0cBERT, \u591a\u6a21\u5f0f\u60c5\u611f\u6aa2\u6e2c \u4e00\u3001\u7c21\u4ecb \u672c\u8ad6\u6587\u91dd\u5c0d Fearless Steps Challenge \u7af6\u8cfd\u4e2d\u7684 sentiment detection \u4efb\u52d9\uff0c\u9032\u884c\u8a9e\u97f3\u60c5 \u611f\u5075\u6e2c\u521d\u6b65\u63a2\u8a0e\u3002Fearless Steps Challenge \u6bd4\u8cfd\uff0c\u662f\u70ba\u4e86\u6176\u795d\u767b\u6708\u8a08\u5283 50 \u9031\u5e74\u6240\u8209\u8fa6\u7684 \u5728\u8a13\u7df4\u7684\u904e\u7a0b\u4e2d\uff0c\u7531 Fearless Steps Challenge \u4f86\u505a SNR \u8cc7\u6599\u5206\u7fa4\uff0cFearless Steps Challenge \u5728\u4f9d\u64da\u5404\u500b\u97f3\u6a94\u7684 SNR(Std Dev)\u6578\u503c\u53bb\u505a\u6d88\u9664\u96dc\u8a0a\u7684\u52d5\u4f5c\u3002 \u4e94\u3001\u5be6\u9a57\u8a9e\u7af6\u8cfd\u7d50\u679c \u672c\u5be6\u9a57\u4f7f\u7528 Fearless Steps Challenge \u8cc7\u6599\u5eab\u884c\u8a13\u7df4\uff0cFearless Steps Challenge \u8cc7\u6599\u5eab \u5206\u6210\u4e86 Train data\u3002\u672c\u5be6\u9a57\u5c07 Train data \u8cc7\u6599\u5eab\u7684\u6bcf\u4efd\u97f3\u6a94\u5229\u7528 Sliding Windows \u7684\u65b9\u5f0f \u5207\u51fa\u8a13\u7df4\u8cc7\u6599\uff0c\u7a97\u53e3\u5927\u5c0f\u70ba 2s \u6bcf\u6b21\u4f4d\u79fb 10ms \u9032\u884c Train data \u7684\u8cc7\u6599\u63a1\u6a23\u3002\u4ee5\u4e0b\u5148\u55ae\u7368 \u5c0d\u5404\u500b\u90e8\u5206\u9032\u884c\u5be6\u9a57\uff0c\u5206\u70ba CNN \u67b6\u69cb\u7684\u97f3\u983b\u90e8\u5206\u548c BERT \u6587\u5b57\u90e8\u5206\u5225\u9032\u884c\u8a0e\u8ad6\uff0c\u518d\u8a0e Sliding Windows \u7684\u65b9\u5f0f\u5207\u51fa\u8a13\u7df4\u8cc7\u6599\uff0c\u7a97\u53e3\u5927\u5c0f\u70ba 2s \u6bcf\u6b21\u4f4d\u79fb 10ms \u9032\u884c\u8a13\u7df4\u3002 \u5728 128 \u8a13\u7df4\u6b21\u6578\u5f8c\uff0c\u6b63\u78ba\u7387\u9054\u5230 60.51%\uff0c \u5728\u7b2c\u4e8c\u6b21\u5be6\u9a57\u4e0b\uff0c\u4fee\u6539 state machine \u7684\u66ab\u5b58\u5668\u500b\u6578\u4f86\u8b93\u60c5\u7dd2\u6d6e\u52d5\u7684\u7bc4\u570d\u4e0d\u6703\u8b8a\u5316\u7684 \u592a\u5feb\uff0c\u5c07 state machine \u66ab\u5b58\u5668\u4fee\u6539\u70ba 15 \u500b\u6700\u672c\u8ad6\u6587\u6700\u9ad8\u6b63\u78ba\u7387\uff0c\u6b63\u78ba\u7387\u4f86\u5230\u4e86 73.11% \u70ba Fearless Steps Challenge \u6bd4\u8cfd\u4e2d Sentiment Detection \u9805\u76ee\u7684 Rank 3 \u6392\u540d\uff0c \u5be6\u9a57\u4e09\uff1a\u63d0\u4ea4\u81f3 Fearless Steps Challenge \u5b98\u65b9\u4e4b\u7cfb\u7d71\u5dee\u7570\u8aaa\u660e \u4ee5\u4e0b\u5206\u5225\u8aaa\u660e\u5728 Fearless Steps Challenge \u5b98\u65b9\u7db2\u7ad9\u7e3d\u6392\u540d\u4e2d\uff0c\u4e0d\u540c NTUT_sys \u7cfb\u7d71 \u7684\u505a\u6cd5\u8207\u8a2d\u5b9a\u5dee\u7570\uff1a 1. NTUT_sys1 \u4f7f\u7528 google \u8a9e\u97f3\u8fa8\u8b58\u5c07\u97f3\u6a94\u5207\u5272\u70ba 15 \u79d2\u4e00\u500b\u55ae\u4f4d\u97f3\u6a94\u4e0d\u91cd\u758a\uff0c\u9032\u884c\u55ae \u97f3\u983b\u6e2c\u8a66\u795e\u7d93\u7db2\u8def\u6a21\u578b\u5982\u5716\u4e09\uff0cFearless Steps Challenge \u5b98\u65b9\u6b63\u78ba\u7387\u70ba 39.49% 2. NTUT_sys2 \u4f7f\u7528 google \u8a9e\u97f3\u8fa8\u8b58\u5c07\u97f3\u6a94\u5207\u5272\u70ba 15 \u79d2\u4e00\u500b\u55ae\u4f4d\u97f3\u6a94\u4e0d\u91cd\u758a\uff0c\u9032\u884c\u591a \u6a21\u8a66\u795e\u7d93\u7db2\u8def\u6a21\u578b\u5982\u5716\u4e00\uff0cFearless Steps Challenge \u5b98\u65b9\u6b63\u78ba\u7387\u70ba 60.51% 3. NTUT_sys3 \u70ba NTUT_sys2 \u7684\u91cd\u8907\u537b\u8a8d\u6b63\u78ba\u7387\uff0c\u56e0\u6b64\u5728\u56de\u50b3\u4e00\u6b21\u7d66 Fearless Steps Challenge \u5b98\u65b9\u537b\u8a8d\u6b63\u78ba\u7387\u70ba 60.51% 4. NTUT_sys4 \u4f7f\u7528 Sliding Windows \u7684\u65b9\u5f0f\u9032\u884c\u9a57\u8b49\u5c07\u97f3\u6846\u8a2d\u70ba 2s \u4f4d\u79fb\u6642\u9593\u70ba 10ms\uff0c \u9032\u884c\u55ae\u97f3\u983b\u6e2c\u8a66\u795e\u7d93\u7db2\u8def\u6a21\u578b\u5982\u5716\u4e09\uff0c Fearless Steps Challenge \u5b98\u65b9\u6b63\u78ba\u7387\u70ba 44.07% 5. NTUT_sys5 \u4f7f\u7528 Sliding Windows \u7684\u65b9\u5f0f\u9032\u884c\u9a57\u8b49\u5c07\u6587\u5b57\u63a1\u6a23\u7bc4\u570d\u8abf\u6574\u70ba\u524d\u5f8c 14 \u500b\u5b57\uff0c\u9032\u884c\u55ae\u6587\u5b57\u6e2c\u8a66\u795e\u7d93\u7db2\u8def\u6a21\u578b\u5982\u5716\u56db\uff0cFearless Steps Challenge \u5b98\u65b9\u6b63\u78ba\u7387 \u70ba 44.63% 6. NTUT_sys6 \u70ba NTUT_ sys4 \u7684\u91cd\u8907\u537b\u8a8d\u6b63\u78ba\u7387\uff0c\u56e0\u6b64\u5728\u56de\u50b3\u4e00\u6b21\u7d66 Fearless Steps Challenge \u5b98\u65b9\u537b\u8a8d\u6b63\u78ba\u7387\u70ba 44.07% 7. NTUT_sys7 \u70ba NTUT_ sys5 \u7684\u91cd\u8907\u537b\u8a8d\u6b63\u78ba\u7387\uff0c\u56e0\u6b64\u5728\u56de\u50b3\u4e00\u6b21\u7d66 Fearless Steps Challenge \u5b98\u65b9\u537b\u8a8d\u6b63\u78ba\u7387\u70ba 44.63% 8. NTUT_sys8 \u4f7f\u7528 Fearless Steps Challenge Train data \u6240\u7b97\u51fa\u7684\u7b54\u6848\u9032\u884c\u56de\u50b3\uff0c\u56e0\u6b64 \u4e0d\u5217\u5728\u5b98\u65b9\u6392\u540d\u4e2d 9. NTUT_sys9 \u4f7f\u7528 Sliding Windows \u7684\u65b9\u5f0f\u9032\u884c\u9a57\u8b49\uff0c\u591a\u6a21\u5f0f\u795e\u7d93\u7db2\u8def\u9032\u884c\u8b58\u5225\uff0c state machine \u66ab\u5b58\u5668\u8a2d\u70ba 3 \u500b\uff0cFearless Steps Challenge \u5b98\u65b9\u6b63\u78ba\u7387\u70ba 70.47% 10. NTUT_sys10 \u4f7f\u7528 Sliding Windows \u7684\u65b9\u5f0f\u9032\u884c\u9a57\u8b49\uff0c\u591a\u6a21\u5f0f\u795e\u7d93\u7db2\u8def\u9032\u884c\u8b58\u5225\uff0c state machine \u66ab\u5b58\u5668\u8a2d\u70ba 15 \u500b\uff0cFearless Steps Challenge \u5b98\u65b9\u6b63\u78ba\u7387\u70ba 73.11% \u516d\u3001\u7d50\u8ad6 \u5728\u672c\u8ad6\u6587\u4e2d\uff0c\u6211\u5011\u63d0\u51fa\u4e86\u57fa\u65bc CNN \u8207 BERT \u7684\u591a\u6a21\u5f0f\u60c5\u7dd2\u8b58\u5225\u795e\u7d93\u7db2\u8def\u67b6\u69cb\uff0c\u878d \u5408\u8072\u5b78\u8207\u8a9e\u610f\u60c5\u7dd2\u7279\u5fb5\u53c3\u6578\uff0c\u7528\u4ee5\u5075\u6e2c\u8a9e\u97f3\u8a0a\u865f\u4e2d\u50b3\u9054\u7684\u60c5\u7dd2\u72c0\u614b\uff0c\u4ee5\u5f37\u5316\u7cfb\u7d71\u7684\u60c5\u7dd2 \u72c0\u614b\u5075\u6e2c\u6548\u80fd\u3002\u4e26\u4ee5 state machine \u6e1b\u7de9\u8f38\u51fa\u8df3\u52d5\u7684\u60c5\u6cc1\uff0c\u6709\u6548\u7684\u89e3\u6c7a\u8f38\u51fa\u6642\u7522\u751f\u7684\u4e0d\u7a69 \u5b9a\u6027\uff0c\u63d0\u5347\u6e96\u78ba\u5ea6\u3002\u6700\u5f8c\uff0c\u7531\u6b63\u5f0f\u6bd4\u8cfd\u7d50\u679c\u767c\u73fe\uff0c\u6211\u5011\u7684\u7cfb\u7d71\u7684\u60c5\u7dd2\u72c0\u614b\u5075\u6e2c\u6b63\u78ba\u7387\u9054 \u5230 73.11%\uff0c\u5728\u6240\u6709\u968a\u4f0d\u63d0\u4ea4\u4e2d\u7684 20 \u500b\u7d50\u679c\u4e2d\uff0c\u6392\u7b2c\u4e09\u540d\uff0c\u4e0d\u4f46\u8d85\u8d8a\u4e3b\u8fa6\u55ae\u4f4d\u63d0\u4f9b\u7684\u57fa \u6e96\u53c3\u8003\u7cfb\u7d71(49.75%) \uff0c\u4e26\u53ea\u5dee\u7b2c\u4e00\u540d(74.07)\u4e0d\u5230 1%\u3002", | |
| "html": null, | |
| "content": "<table><tr><td>\u56e0\u6b64\uff0c\u672c\u8ad6\u6587\u63d0\u51fa\u5982\u5716\u4e00\u7684\u591a\u6a21\u5f0f\u60c5\u7dd2\u5075\u6e2c\u6a21\u578b\u3002\u4e3b\u8981\u60f3\u6cd5\u662f\u540c\u6642\u8003\u616e\u8a9e\u97f3\u8a0a\u865f\u4e2d 25 \u5c0f\u6642\uff0c\u767b\u6708\u7d04 50 \u5c0f\u6642\uff0c\u6708\u7403\u6f2b\u6b65\u7d04 25 \u5c0f\u6642\u3002\u6b64\u5916\u7531\u65bc\u4efb\u52d9\u7684\u4e0d\u540c\uff0c\u8a9e\u6599\u5eab\u7684\u8a9e\u97f3 \u5728 neutral \u7684\u5224\u65b7\u9084\u662f\u6709\u90e8\u5206\u4e9b\u8a31\u4e0d\u6e96\u78ba\u3002 \u672c\u8ad6\u6587\u5c07\u6587\u5b57\u55ae\u7368\u4f7f\u7528 BERT \u795e\u7d93\u7db2\u8def\u6a21\u578b\u9032\u884c\u55ae\u7368\u8cc7\u6599\u5eab\u8a13\u7df4\uff0c\u6a21\u578b\u5982\u5716\u56db\uff0c\u5728</td></tr><tr><td>\u5305\u542b\u7684\u8072\u5b78\u8207\u8a9e\u610f\u8cc7\u8a0a\uff0c\u63d0\u51fa\u57fa\u65bc\u6df1\u5ea6\u985e\u795e\u7d93\u7db2\u8def\u4e4b\u591a\u6a21\u5f0f\u8a9e\u97f3\u60c5\u7dd2\u5075\u6e2c\u6a21\u578b\uff0c\u7528\u4ee5\u5075 \u6d3b\u52d5\u5bc6\u5ea6\u5728\u6574\u500b\u4efb\u52d9\u4e2d\u5e38\u5e38\u8b8a\u5316\uff0c\u4e14\u8a9e\u97f3\u6578\u64da\u7684\u8cea\u91cf\u4e5f\u5e38\u5728 0 \u5230 20dB (Signal-to-noise \u672c\u8ad6\u6587\u5c07\u97f3\u983b\u55ae\u7368\u4f7f\u7528 CNN \u795e\u7d93\u7db2\u8def\u6a21\u578b\u9032\u884c\u55ae\u7368\u8cc7\u6599\u5eab\u8a13\u7df4\uff0c\u6a21\u578b\u5982\u5716\u4e09\uff0c\u5728 Fearless Steps Challenge \u5b98\u65b9\u7db2\u7ad9\u6e96\u78ba\u7387\u70ba 44.63%\u53c3\u8003\u6392\u540d\u5982\u8868\u4e09\uff0c\u56e0\u6b64\u4f7f\u7528\u55ae\u6587\u5b57\u8b58</td></tr><tr><td>\u6e2c\u8a9e\u97f3\u8a0a\u865f\u4e2d\u50b3\u9054\u7684\u60c5\u7dd2\u72c0\u614b\u3002\u5be6\u969b\u505a\u6cd5\u5305\u62ec(1)\u5229\u7528\u6372\u7a4d\u795e\u7d93\u7db2\u7d61(Convolutional Neural Network, CNN) \uff0c\u5f9e\u8072\u5b78\u983b\u8b5c\u81ea\u52d5\u6c42\u53d6\u60c5\u7dd2\u7279\u5fb5\u53c3\u6578\uff0c\u8207(2)\u4ee5\u96d9\u5411\u7de8\u78bc\u8b8a\u63db\u5668 ratio , SNR)\u4e4b\u9593\u8b8a\u5316\u3002Fearless Steps Challenge \u70ba\u4e86\u78ba\u4fdd\u80fd\u5c07\u6578\u64da\u516c\u5e73\u5730\u5206\u914d\u5230\u7684\u8a13\u7df4\uff0c \u8a55\u4f30\u548c\u958b\u767c\u5b50\u96c6\u4e2d\uff0c\u6839\u64da\u566a\u8072\u6c34\u5e73\u8207\u6d3b\u52d5\u5bc6\u5ea6\uff0c\u5c0d\u6578\u64da\u9032\u884c\u5206\u985e\u3002 \u5728\u9577\u6642\u8a13\u7df4\u53ca\u8fa8\u8b58\u7684\u6587\u5b57\u548c\u8a9e\u97f3\u7531 Fearless Steps Challenge \u6240\u63d0\u4f9b\u7684\u7f8e\u570b\u5b87\u822a\u5c40\u963f \u5225\u60c5\u7dd2\u6642\u5982\u679c\u6709\u76f8\u95dc\u60c5\u7dd2\u5b57\u773c\u51fa\u73fe\u6642\u5247\u6703\u5c0d\u6e96\u78ba\u7387\u7167\u6210\u4e00\u5b9a\u7684\u5f71\u97ff\uff0c\u4f46\u67d0\u4e9b\u5834\u666f\u4e0b\u7121\u6cd5 Fearless Steps Challenge \u5b98\u65b9\u7db2\u7ad9\u6e96\u78ba\u7387\u70ba 44.07%\u53c3\u8003\u6392\u540d\u5982\u8868\u4e94\uff0c\u56e0\u6b64\u55ae\u97f3\u983b\u6e2c\u8a66\u5c0d \u4e09\u3001\u57fa\u65bc\u591a\u6a21\u5f0f\u4e4b\u60c5\u7dd2\u6aa2\u6e2c\u7cfb\u7d71 \u6ce2\u7f85\u8a08\u5283\u7684\u5168\u7a0b\u7121\u7dda\u96fb\u8cc7\u6599\u5eab\u88e1\uff0c\u6211\u5011\u4f7f\u7528\u6df7\u5408\u6a21\u578b\u5c07\u8a9e\u97f3\u53ca\u6587\u5b57\u9032\u884c\u540c\u6b65\u8a13\u7df4\u3002 \u65bc silence \u548c neutral \u5075\u6e2c\u6709\u4e00\u5b9a\u7684\u6e96\u78ba\u5ea6\uff0c\u4f46\u96e2\u6700\u9ad8\u6e96\u78ba\u7387\u9084\u662f\u9700\u8981\u9760\u6587\u5b57\u7684\u8f14\u52a9\u4e0b\u9054 \u7d14\u7cb9\u4f9d\u9760\u55ae\u6587\u5b57\u6e2c\u8a66\uff0c\u56e0\u6b64\u672c\u8ad6\u6587\u4f7f\u7528\u591a\u6a21\u5f0f\u795e\u7d93\u7db2\u8def\u6a21\u578b\u5c07\u5169\u8005\u6a21\u578b\u6df7\u5408\u3002</td></tr><tr><td>\u5927\u898f\u6a21\u7af6\u8cfd\u3002\u7531\u5fb7\u5dde\u8cfd\u62c9\u9054\u5206\u6821\u5c07\u767b\u6708\u4efb\u52d9\u4e2d\u6240\u6709\u901a\u8a0a\u5c0d\u8a71\u6578\u4f4d\u5316\uff0c\u4e26\u767c\u884c Fearless Steps Corpus \u8a9e\u6599\uff0c\u652f\u63f4\u5404\u500b\u7af6\u8cfd\u9805\u76ee\uff0c\u63d0\u4f9b\u5927\u91cf\u7684\u8a13\u7df4\u8cc7\u6599\u53ca\u6e2c\u8a66\u8cc7\u6599\uff0c\u56e0\u70ba\u6b64\u9805\u6bd4 \u8cfd\u4e3b\u8981\u662f\u5e0c\u671b\u53ef\u4ee5\u4f7f\u7528\u81ea\u7136\u74b0\u5883\u7576\u4e2d\u6240\u9304\u88fd\u7684\u8cc7\u6599\u5eab\u9032\u884c\u6bd4\u8cfd\uff0c\u6240\u4ee5 Fearless Steps (Bidirectional Encoder Representation from Transformers, BERT) \uff0c\u6c42\u53d6\u8a9e\u97f3\u9010\u5b57\u7a3f\u7684\u8a9e Fearless Steps Challenge \u6240\u63d0\u4f9b\u7684\u8a13\u7df4\u5b50\u96c6\uff0c\u7686\u7d93\u4eba\u5de5\u8f49\u5beb\u9010\u5b57\u7a3f\u8207\u6a19\u8a18\u60c5\u7dd2\u6a19\u7c64\u3002 \u7279\u5fb5\u7d1a\u878d\u5408\u662f\u6700\u5e38\u898b\u548c\u76f4\u63a5\u7684\u65b9\u5f0f\uff0c\u5176\u4e2d\u6240\u6709\u63d0\u53d6\u7684\u7279\u5fb5\u76f4\u63a5\u9023\u63a5\u6210\u55ae\u500b\u9ad8\u7dad\u7279\u5fb5 \u6210\u3002 \u5be6\u9a57\u4e8c\uff0c\u591a\u6a21\u5f0f\u60c5\u7dd2\u5075\u6e2c (\u4e00)\u3001\u5377\u7a4d\u795e\u7d93\u7db2\u8def\u8072\u5b78\u60c5\u7dd2\u6a21\u578b\u67b6\u69cb \u610f\u7279\u5fb5\u53c3\u6578\u3002\u518d\u5c07\u6b64\u5169\u985e\u7279\u5fb5\u53c3\u6578\u5411\u91cf\u878d\u5408\uff0c\u4ee5\u5f37\u5316\u7cfb\u7d71\u7684\u60c5\u7dd2\u72c0\u614b\u5075\u6e2c\u6548\u80fd\u3002\u5716\u4e00\u70ba \u6211\u5011\u9032\u884c\u60c5\u7dd2\u5075\u6e2c\u7684\u6846\u67b6\u7d50\u69cb\uff1a \u8a55\u4f30\u5b50\u96c6\u5247\u53ea\u63d0\u4f9b\u81ea\u52d5\u7522\u751f\u7684\u9010\u5b57\u7a3f\u8207\u60c5\u7dd2\u6a19\u7c64\u3002\u4f46 \u6e2c\u8a66\u5b50\u96c6\u5247\u7121\u63d0\u4f9b\u4efb\u4f55\u60c5\u7dd2\u6a19\u7c64\u4e5f \u7121\u9010\u5b57\u7a3f\uff0c\u56e0\u6b64\u672c\u8ad6\u6587\u5728\u6e2c\u8a66\u8cc7\u6599\u6642\uff0c\u9700\u8981\u5c0d\u97f3\u6a94\u5148\u9032\u884c\u6587\u5b57\u8f49\u5beb\u8655\u7406\u3002\u7531\u65bc Fearless Steps Challenge \u6240\u63d0\u4f9b\u5f97\u662f\u592a\u7a7a\u4e2d\u4e0d\u540c\u5834\u666f\u7684\u8a9e\u97f3\u8a18\u9304\uff0c\u7e3d\u5171\u63d0\u4f9b\u4e86\u4e94\u500b\u4e0d\u540c\u90e8\u5206\u7684\u983b \u672c\u7bc7\u7896\u6587\u63d0\u51fa\u7684\u65b9\u6cd5\u7684\u7b2c\u4e00\u968e\u6bb5\uff0c\u5148\u5c0d\u8f38\u5165\u8a9e\u97f3\u4fe1\u865f\u57f7\u884c\u53d6\u97f3\u6846\u8207\u6c42\u53d6\u8a9e\u97f3\u4fe1\u865f\u7684 \u983b\u8b5c\u3002\u5176\u4e2d\u6211\u5011\u4f7f\u7528 2 \u79d2\u7684\u7a97\u53e3\u5927\u5c0f\u8207 10ms \u7684\u97f3\u6846\u4f4d\u79fb\uff0c\u4f86\u7372\u5f97\u8db3\u5920\u53ef\u8a13\u7df4\u8cc7\u6599\u3002\u7136 \u5716\u4e09\u3001CNN Architecture for Sentiment Detection \u5411\u91cf\u3002\u7136\u5f8c\uff0c\u53ef\u4ee5\u7528\u9019\u7a2e\u9ad8\u7dad\u7279\u5fb5\u5411\u91cf\u8a13\u7df4\u55ae\u500b\u5206\u985e\u5668\u7528\u65bc\u60c5\u7dd2\u8b58\u5225\u3002\u5927\u91cf\u5148\u524d\u7684\u4f5c\u54c1 \u8ad6\u591a\u6a21\u5f0f\u60c5\u7dd2\u5075\u6e2c\u6a21\u578b\u3002 2. \u6587\u5b57 BERT \u5728\u591a\u6a21\u5f0f\u5be6\u9a57\u7576\u4e2d\u672c\u8ad6\u6587\u4f7f\u7528\u97f3\u983b\u6a21\u578b\u548c\u6587\u5b57\u6a21\u578b\u9032\u884c\u60c5\u7dd2\u8b58\u5225\uff0c\u4ea4\u53c9\u6e2c\u8a66\u767c\u73fe\u97f3 [15-19]\u8b49\u660e\u4e86\u60c5\u611f\u8b58\u5225\u4efb\u52d9\u4e2d\u7279\u5fb5\u7d1a\u878d\u5408\u7684\u8868\u73fe\u3002\u4f46\u662f\uff0c\u56e0\u70ba\u5b83\u4ee5\u76f4\u63a5\u7684\u65b9\u5f0f\u5408\u4f75\u97f3\u983b \u548c\u6587\u5b57\u7279\u5fb5\uff0c\u6240\u4ee5\u7279\u5fb5\u7d1a\u878d\u5408\u4e0d\u80fd\u6a21\u64ec\u8907\u96dc\u7684\u95dc\u4fc2\u3002 \u5177\u9ad4\u5730\uff0c\u6bcf\u500b\u8f38\u5165\u6a21\u614b\u7528\u60c5\u7dd2\u5206 \u6b64\u5916\uff0c\u5716\u516d\u70ba\u6211\u5011\u63d0\u4ea4\u81f3 Fearless Steps Challenge \u5b98\u65b9\uff0c\u7d93\u904e\u5b98\u65b9\u8a55\u6e2c\u5f8c\u7684\u6210\u7e3e\u6392 \u7531\u65bc Fearless Steps Challenge \u6e2c\u8a66\u96c6\u4e26\u7121\u63d0\u4f9b\u97f3\u983b\u7684\u6587\u5b57\uff0c\u56e0\u6b64\u5728\u4f7f\u7528\u6e2c\u8a66\u96c6\u8fa8\u8b58 \u983b\u6e2c\u8a66\u7576\u4e2d\u767c\u73fe positive \u548c negative \u7684\u60c5\u7dd2\u985e\u5225\u8f03\u70ba\u4e0d\u6e96\u78ba\uff0c\u6587\u5b57\u6e2c\u8a66\u90e8\u5206\u4e5f\u767c\u73fe\u6587\u5b57 \u5716\u4e03\u3001state machine \u72c0\u614b\u5716</td></tr><tr><td>Corpus \u7684\u8a9e\u97f3\u8cc7\u6599\uff0c\u662f\u771f\u5be6\u592a\u7a7a\u4efb\u52d9\u4e2d\uff0c\u592a\u7a7a\u4eba\u8207\u4efb\u52d9\u4e2d\u5fc3\u7684\u901a\u8a0a\u5c0d\u8a71\u9304\u97f3\u3002 \u6211\u5011\u6703\u9078\u64c7\u53c3\u52a0\u6b64\u9805\u7af6\u8cfd\uff0c\u4e3b\u8981\u662f\u56e0\u70ba\u76ee\u524d\u5927\u90e8\u5206\u53ef\u53d6\u5f97\u7684\u60c5\u7dd2\u76f8\u95dc\u8a9e\u6599\u5eab\uff0c\u5927\u90fd \u9053\u5834\u666f\uff0cFlight Director (FD)\u3001Mission Operations Control Room (MOCR) \u3001Guidance Navigation and Control (GNC)\u3001Network Controller (NTWK)\u3001Electrical, Environmental and \u5f8c\u5c07\u8a0a\u865f\u8f49\u63db\u81f3\u983b\u57df\uff0c\u5728\u6b64\u8655\u4ee5 Mel-frequency \u4e09\u89d2\u5f62\u6ffe\u6ce2\u5668\u7d44\u904e\u6ffe\u983b\u8b5c\uff0c\u8f49\u6210 Mel-frequency filterbank \u53c3\u6578\uff0c\u518d\u7372\u5f97\u6700\u7d42\u7684 MFCCs\u3002\u5728\u672c\u6587\u4e2d\uff0c\u6211\u5011\u9084\u4f7f\u7528 20 \u500b\u6ffe\u6ce2\u5668 (\u4e8c)\u3001BERT \u795e\u7d93\u7db2\u8def\u8a9e\u610f\u60c5\u7dd2\u6a21\u578b\u67b6\u69cb \u985e\u5668\u7368\u7acb\u5efa\u6a21\uff0c\u7136\u5f8c\u5c07\u9019\u4e9b\u8b58\u5225\u7d50\u679c\u8207\u67d0\u4e9b\u4ee3\u6578\u898f\u5247\u7d44\u5408\uff0c\u4f8b\u5982\uff1a\"Max\"\u3001\"Min\"\u3001\"Sum\" \u7b49\u3002\u56e0\u6b64\uff0c\u5728\u60c5\u611f\u8b58\u5225\u4e2d\u63a1\u7528\u4e86\u6c7a\u7b56\u878d\u5408\u3002\u7136\u800c\uff0c\u6c7a\u7b56\u5c64\u878d\u5408\u7121\u6cd5\u6355\u6349\u4e0d\u540c\u6a21\u614b\u4e4b\u9593\u7684 \u6642\u5c07\u97f3\u983b\u4f7f\u7528\u8a9e\u97f3\u8fa8\u8b58(Automatic Speech Recognition, ASR)\u9032\u884c\u8fa8\u8b58\uff0c\u4f46\u8a9e\u97f3\u8fa8\u8b58\u53ea\u80fd \u63a1\u6a23\u7bc4\u570d\u5c0d\u65bc\u5be6\u9a57\u7d50\u679c\u6709\u4e00\u5b9a\u5f71\u97ff\uff0c\u97f3\u983b\u7121\u6cd5\u8b58\u5225\u7684 positive \u548c negative \u5229\u7528\u591a\u6a21\u5f0f\u6a21 \u540d\u7d50\u679c\uff0c\u6211\u5011\u7e3d\u5171\u63d0\u4ea4\u4e86 10 \u500b\u4e0d\u540c\u8a2d\u5b9a\u7684\u7cfb\u7d71\uff0c\u5728\u4ee5\u4e0b\u5be6\u9a57\u4e2d\u6703\u9010\u6b65\u8aaa\u660e\u3002 (\u4e8c) \u8a55\u4f30\u6307\u6a19 \u7372\u5f97\u55ae\u8a5e\u8d77\u59cb\u6642\u9593\u548c\u7d50\u675f\u6642\u9593\uff0c\u6240\u4ee5\u5728\u6587\u5b57\u9810\u8655\u7406\u672c\u8ad6\u6587\u4f7f\u7528 Sliding Windows \u65b9\u5f0f\uff0c \u578b\uff0c\u7531\u6587\u5b57\u6a21\u578b\u4f86\u8b58\u5225 positive \u548c negative \u7684\u76f8\u95dc\u60c5\u7dd2\u5b57\u773c\uff0c\u4ee5\u5229\u65bc\u63d0\u5347\u6a21\u578b\u6e96\u78ba\u7387\uff0c</td></tr><tr><td>\u662f\u7531\u6f14\u54e1\u8868\u6f14\u7684\uff0c\u4e14\u8a9e\u6599\u90fd\u904e\u65bc\u5b8c\u7f8e\u6216\u662f\u904e\u65bc\u4e7e\u6de8\uff0c\u5c0e\u81f4\u5728\u9019\u4e9b\u8a9e\u6599\u5eab\u4e0a\u6240\u7372\u5f97\u7684\u7d50\u679c\uff0c Consumables Manager (EECOM)\uff0c\u8868\u4e00\u63d0\u4f9b\u4e0d\u540c\u4e8b\u4ef6\u7684\u4e94\u500b\u5834\u666f\u7684\u6642\u9593\u5206\u5e03\u8868\u3002 \u7d44\u548c 40-MFCC \u9032\u884c\u7279\u5fb5\u63d0\u53d6\uff0c\u518d\u5c07 MFCCs \u77e9\u9663\u8f38\u5165 CNN \u4e2d\u3002 \u76f8\u4e92\u95dc\u806f\uff0c\u56e0\u70ba\u9019\u4e9b\u6a21\u614b\u88ab\u8a8d\u70ba\u662f\u7368\u7acb\u7684\u3002\u56e0\u6b64\uff0c\u6c7a\u7b56\u7d1a\u878d\u5408\u4e0d\u7b26\u5408\u4eba\u985e\u60c5\u7dd2\u7279\u5fb5\u7684\u7279 Fearless Steps Challenge \u6bd4\u8cfd\u898f\u5247\u5982\u4e0b\uff0c\u97f3\u6a94\u5be6\u969b\u5224\u65b7\u6b63\u78ba\u6642\u9593\u9577\u5ea6\u55ae\u4f4d\u70ba 10ms\uff0c \u5728\u8981\u8fa8\u8b58\u55ae\u8a5e\u6642\u9593\u5e95\u4e0b\u5f80\u524d\u5f80\u5f8c\u53d6\u4e00\u5b9a\u7bc4\u570d\u7684\u55ae\u8a5e\u91cf\u7d44\u6210\u53e5\u5411\u91cf\uff0c\u8f38\u5165\u5982\u5716\u56db\u6240\u793a \u97f3\u983b\u6e2c\u8a66\u7576\u4e2d silence \u548c neutral \u7684\u6e96\u78ba\u5ea6\u4e5f\u6709\u4e00\u5b9a\u6210\u6548\uff0c\u56e0\u6b64\u4e5f\u53ef\u8f14\u52a9\u591a\u6a21\u578b\u8b58\u5225 Non-</td></tr><tr><td>\u4e0d\u898b\u5f97\u53ef\u4ee5\u53cd\u61c9\u5be6\u969b\u904b\u7528\u6642\u7684\u60c5\u5883\u3002\u800c Fearless Steps Challenge \u7684\u8a9e\u97f3\u8cc7\u6599\uff0c\u662f\u771f\u5be6\u592a \u7a7a\u4efb\u52d9\u4e2d\uff0c\u592a\u7a7a\u4eba\u8207\u4efb\u52d9\u4e2d\u5fc3\u7684\u901a\u8a0a\u5c0d\u8a71\u9304\u97f3\uff0c\u56e0\u6b64\u6703\u6709\u8a31\u591a\u81ea\u7136\u7684\u96dc\u8a0a\u548c\u5c0d\u8a71\u3002\u6700\u91cd \u8981\u7684\u662f\uff0c\u6b64 Fearless Steps Corpus \u8a9e\u6599\u5eab\u7e3d\u5171\u5305\u542b 100 \u5c0f\u6642\u7684\u8a9e\u6599\uff0c\u800c\u4e14\u60c5\u7dd2\u6a19\u7c64\u90fd\u662f \u5728\u53c3\u8003\u7b54\u6848\u7576\u4e2d\u53ea\u6709\u5075\u6e2c\u5230\u548c\u53c3\u8003\u7b54\u6848\u7bc4\u570d\u5167\u4e00\u6a23\u624d\u7d66\u4e88\u5f97\u5206\u5982\u5716\u4e94\uff0c\u82e5\u5224\u65b7\u8d85\u51fa\u53c3\u8003 BERT \u6a21\u578b\u5167\u9032\u884c\u55ae\u6587\u5b57\u6e2c\u8a66\u3002 Sentiment \u7684\u5207\u5272\u4f4d\u7f6e\u6e96\u78ba\u5ea6\u548c neutral \u7684\u6b63\u78ba\u7387\uff0c\u6240\u4ee5\u672c\u8ad6\u6587\u4f7f\u7528\u6df7\u5408\u795e\u7d93\u7db2\u8def\u6a21\u578b\u67b6 \u6027\u3002 \u8868\u4e00\u3001Total Speech Durations per Channel and Event ECOM FD GNC MOCR NTWK \u6a21\u578b\u7d1a\u878d\u5408\u4f5c\u70ba\u7279\u5fb5\u7d1a\u878d\u5408\u548c\u6c7a\u7b56\u7d1a\u878d\u4e4b\u9593\u7684\u6298\u8877\uff0c\u4e5f\u88ab\u7528\u65bc\u60c5\u611f\u8b58\u5225\u6700\u4f73\u89e3\u6c7a\u65b9 \u69cb\u4f86\u63d0\u5347\u6a21\u578b\u6e96\u78ba\u5ea6\u3002 \u5f97\u5206\u7bc4\u570d\u5247\u4e0d\u6263\u5206\u4e5f\u4e0d\u4e88\u8a08\u5206\u53ea\u8a08\u7b97\u771f\u5be6\u5f97\u5206\u6578\uff0c\u6bcf\u500b\u5f97\u5206\u5340\u57df\u5c07\u8a08\u7b97\u6bcf 10ms \u5e40\u7684\u771f \u5728\u6e2c\u8a66\u6587\u5b57\u4e2d\u767c\u73fe\uff0c\u6587\u5b57\u63a1\u6a23\u7bc4\u570d\u4e0d\u540c\u6642\u6703\u6709\u4e0d\u540c\u6e96\u78ba\u7387\uff0c\u7576\u63a1\u6a23\u6587\u5b57\u63a1\u6a23\u7bc4\u570d\u9054 Total Lift Off 2.1 1.2 1.3 0.8 3.9 9.3 \u6cd5\u3002\u8a72\u65b9\u6cd5\u65e8\u5728\u7372\u5f97\u97f3\u983b\u548c\u6587\u5b57\u6a21\u614b\u7684\u806f\u5408\u7279\u5fb5\u8868\u793a\u3002\u5176\u5be6\u73fe\u4e3b\u8981\u53d6\u6c7a\u65bc\u6240\u4f7f\u7528\u7684\u878d\u5408 \u5be6\u76f8\u540c\u7b54\u6848\u7684\u6578\u503c(\u6a19\u7c64\u4e0a\u7684\u6700\u4f4e\u5206\u8fa8\u7387) \u3002 \u5230\u5f80\u524d\u5f80\u5f8c 14 \u500b\u5b57\u6642\u4e4b\u5f8c\u6e96\u78ba\u7387\u8da8\u8fd1\u65bc\u7a69\u5b9a\uff0c\u5982\u8868\u56db\u6240\u793a\uff0c\u5728\u63a1\u6a23\u7bc4\u570d\u5f9e 2 \u81f3 8 \u500b\u5b57 \u8cc7\u6599\u5eab\u5206\u70ba\u4e09\u985e negative\uff0cneutral\uff0cpositive\uff0c\u9032\u884c\u9019\u4e09\u985e\u7684\u8fa8\u5225\u3002\u56e0\u60c5\u7dd2\u8b8a\u5316\u52d5\u614b</td></tr><tr><td>\u7d93\u7531\u4eba\u5de5\u6a19\u8a3b\u9a57\u8b49\uff0c\u56e0\u6b64\u7814\u7a76\u7372\u5f97\u7684\u7d50\u679c\u6703\u66f4\u52a0\u6709\u516c\u4fe1\u529b\u3002 Lunar Landing 3.7 1.3 4.0 0.9 \u6a21\u578b\u3002\u4f8b\u5982\uff0c[4]\u63a1\u7528(Hidden Markov Model , MFHMM)\u4f86\u5be6\u73fe\u6a21\u578b\u7d1a\u878d\u5408\u3002 [8]\u63a1\u7528\u8aa4 4.4 14.3 \u6642\u660e\u986f\u63a1\u6a23\u7279\u5fb5\u4e0d\u8db3\u56e0\u6b64\u9020\u6210\u6e96\u78ba\u7387\u6c92\u6709\u660e\u986f\u63d0\u5347\uff0c\u56e0\u6b64\u5c07\u63a1\u6a23\u7bc4\u570d\u63d0\u5347\u5f9e 9 \u81f3 22 \u500b \u8f03\u6162\u800c\u6211\u5011\u6240\u4f7f\u7528\u7684 Sliding Windows \u7684\u8fa8\u8b58\u65b9\u5f0f\u8b93\u7d50\u679c\u8f38\u51fa\u7684\u8b8a\u5316\u592a\u5927\uff0c\u6240\u4ee5\u5728\u9023\u7e8c</td></tr><tr><td>\u50b3\u7d71\u4e0a\u91dd\u5c0d\u8a9e\u97f3\u60c5\u7dd2\u5075\u6e2c\uff0c\u901a\u5e38\u5c08\u6ce8\u65bc\u5148\u63d0\u53d6\u60c5\u7dd2\u7684\u4f4e\u968e\u8072\u5b78\u7279\u5fb5[3-14]\u3002\u4e00\u4e9b\u5ee3 \u6cdb\u4f7f\u7528\u7684\u983b\u8b5c\u7279\u5fb5\u662f Mel-Frequency Cepstral Coefficients(MFCC)[1]\u3001\u7dda\u6027\u9810\u6e2c\u5012\u8b5c\u4fc2 \u6578\u6216\u662f\u97f3\u9ad8\u8ecc\u8de1\u3002\u7136\u5f8c\u518d\u7528\u9ad8\u65af\u6df7\u5408\u6a21\u578b\u3001\u652f\u6301\u5411\u91cf\u6a5f\u6216\u662f\u99ac\u723e\u53ef\u592b\u6a21\u578b\u9032\u884c\u60c5\u7dd2\u8fa8\u8a8d\u3002 \u5716\u4e00\u3001\u591a\u6a21\u5f0f\u795e\u7d93\u7db2\u8def\u6a21\u578b\u67b6\u69cb\u5716 \u6b64\u7cfb\u7d71\u7684\u904b\u4f5c\u5305\u542b\u4e09\u500b\u6a21\u7d44\uff0c\u5305\u62ec\uff1a( 1)\u6211\u5011\u5c07\u539f\u59cb\u8a9e\u97f3\u4fe1\u865f\u8f49\u63db\u70ba\u985e\u4f3c\u5716\u50cf\u7684\u983b Lunar Walking 3.9 1.1 3.0 1.4 2.8 12.2 Total 9.7 3.6 8.3 3.1 11.1 \u5b57\u9032\u884c\u6e2c\u8a66\uff0c\u5728 8 \u81f3 9 \u500b\u5b57\u6642\u6e96\u78ba\u7387\u6709\u660e\u986f\u63d0\u5347\uff0c\u7531\u6b64\u5be6\u9a57\u53ef\u8b49\u5be6\u7576\u6587\u5b57\u63a1\u6a23\u7bc4\u570d\u6703\u5c0d \u8fa8\u8a8d\u6642\u8a2d\u7f6e\u4e86 state machine \u7684\u8f38\u51fa\u6a5f\u5236\uff0c\u5728\u9023\u7e8c\u8f38\u51fa\u4e00\u5b9a\u6578\u91cf\u7684\u7b54\u6848\u624d\u6703\u78ba\u5b9a\u8f38\u51fa\u5426\u5247 \u5dee\u534a\u8026\u5408\u99ac\u723e\u53ef\u592b\u6a21\u578b\u878d\u5408\u4ee5\u9032\u884c\u60c5\u611f\u8b58\u5225\u3002\u5c0d\u65bc\u795e\u7d93\u7db2\u7d61\uff0c\u901a\u904e\u9996\u5148\u9023\u63a5\u5c0d\u61c9\u65bc\u591a\u500b 35.8 \u65bc\u60c5\u7dd2\u8b58\u5225\u6e96\u78ba\u7387\u6709\u4e00\u5b9a\u7684\u6210\u6548\u3002 \u6703\u7e7c\u7e8c\u8f38\u51fa\u524d\u4e00\u500b\u7b54\u6848\u3002\u4f8b\u5982\uff1a\u5716\u516b\u5728\u672a\u4f7f\u7528 state machine \u6642\u8f38\u51fa\u7b54\u6848\u4e0d\u7a69\u5b9a\u6703\u4e00\u76f4\u8df3 \u8f38\u5165\u6a21\u614b\u7684\u795e\u7d93\u7db2\u7d61\u7684\u4e0d\u540c\u96b1\u85cf\u5c64\u7684\u7279\u5fb5\u8868\u793a\u4f86\u57f7\u884c\u6a21\u578b\u7d1a\u878d\u5408\u3002\u7136\u5f8c\uff0c\u6dfb\u52a0\u984d\u5916\u7684\u96b1 \u5716\u4e8c\u3001Sliding Windows for Sentiment Detection \u793a\u610f\u5716 \u8868\u56db\u3001BERT \u6a21\u578b\u5404\u7a2e\u6587\u5b57\u63a1\u6a23\u7bc4\u570d\u6b63\u78ba\u7387 \u85cf\u5c64\u4ee5\u5f9e\u9023\u63a5\u7684\u7279\u5fb5\u5b78\u7fd2\u806f\u5408\u7279\u5fb5\u8868\u793a\u3002\u73fe\u6709\u7684\u6a21\u578b\u7d1a\u878d\u5408\u65b9\u6cd5\u4ecd\u7136\u4e0d\u80fd\u6709\u6548\u5730\u6a21\u64ec\u97f3 \u52d5\uff0c\u4f46\u5728\u52a0\u4e0a state machine \u5f8c\u53ef\u4ee5\u770b\u5230\u7b54\u6848\u8f38\u51fa\u8da8\u8fd1\u65bc\u7a69\u5b9a\u3002</td></tr><tr><td>\u800c\u82e5\u60f3\u5f9e\u8a9e\u610f\u4f86\u6c42\u53d6\u60c5\u7dd2\u7279\u5fb5\u53c3\u6578\uff0c\u5247\u9700\u8981\u5148\u6709\u8a9e\u97f3\u8fa8\u8a8d\u5668\uff0c\u5c07\u8a9e\u97f3\u8f49\u6210\u9010\u5b57\u7a3f\uff0c\u518d\u4ee5 \u81ea\u7136\u8a9e\u8a00\u8655\u7406\u65b9\u5f0f\uff0c\u4f8b\u5982\u4ee5 word-to-vector \u6c42\u53d6\u7279\u5fb5\u5411\u91cf\uff0c\u518d\u4ee5\u985e\u795e\u7d93\u7db2\u8def\u9032\u884c\u60c5\u7dd2\u8fa8 \u8a8d\u3002\u7136\u800c\uff0c\u4eba\u985e\u60c5\u611f\u8207\u8072\u5b78\u4f4e\u968e\u7279\u5fb5\u7684\u8868\u73fe\uff0c\u5be6\u969b\u4e0a\u4e0d\u898b\u5f97\u4e00\u81f4\u3002\u800c\u82e5\u7528\u9010\u5b57\u7a3f\uff0c\u5247\u901a \u5e38\u6703\u6709\u8a9e\u97f3\u8fa8\u8a8d\u932f\u8aa4\uff0c\u5f71\u97ff\u6700\u7d42\u5224\u65b7\u7684\u60c5\u5f62\u3002 \u91dd\u5c0d Fearless Steps Challenge \u6bd4\u8cfd\uff0c\u6211\u5011\u5728\u9032\u884c\u521d\u6b65\u5be6\u9a57\u6e2c\u8a66\u6642\uff0c\u767c\u73fe\u82e5\u55ae\u7368\u53ea\u7528 \u8072\u97f3\u88fd\u4f5c\u6a21\u578b\uff0c\u6216\u662f\u55ae\u7368\u4f7f\u7528\u6587\u5b57\u8a13\u7df4\u6a21\u578b\uff0c\u6240\u5f97\u5230\u6548\u679c\u90fd\u6709\u6240\u4e0d\u8db3\u3002\u4e3b\u8981\u662f\u8a9e\u97f3\u4e2d\u7684 \u5728\u6211\u5011\u7684\u4f8b\u5b50\u4e2d\uff0cCNN \u626e\u6f14\u4e00\u500b\u5f9e\u8a9e\u97f3\u8a0a\u865f\u983b\u8b5c\u4e2d\uff0c\u63d0\u53d6\u8072\u5b78\u60c5\u7dd2\u7279\u5fb5\u53c3\u6578\u7684\u91cd\u8981 \u983b\u548c\u6587\u5b57\u6a21\u614b\u4e4b\u9593\u7684\u9ad8\u5ea6\u975e\u7dda\u6027\u76f8\u95dc\u6027\u3002 state machine \u72c0\u614b\u5716\u5982\u5716\u4e03\uff0c\u9810\u8a2d\u8f38\u51fa\u70ba 0 \u7576 D \u9023\u7e8c\u8f38\u51fa\u4e09\u6b21\u8f49\u614b\u70ba 1 \u6642 VAD \u624d \u8b5c\u5716\u65b9\u5f0f\uff0c\u4f5c\u70ba CNN \u7684\u8f38\u5165[2]\u3002\u56e0\u6b64\uff0c\u53ef\u4ee5\u4f7f\u7528\u4ee5\u5927\u91cf\u8a9e\u97f3\u8a9e\u6599\u9810\u8a13\u7df4\u7684\u6df1\u5ea6 CNN \u6a21\u578b\u9032\u884c\u5b78\u7fd2\uff0c\u64f7\u53d6\u9ad8\u7d1a\u8072\u5b78\u60c5\u7dd2\u7279\u5fb5\u3002 (2)\u5c0d\u65bc\u9010\u5b57\u7a3f\u7684\u591a\u500b\u9023\u7e8c\u6bb5\u843d\uff0c\u53ef\u4ee5\u7528\u4ee5\u5927 \u898f\u6a21\u6587\u5b57\u6578\u64da\u96c6\u9810\u8a13\u7df4\u7684 BERT \u6a21\u578b\u9032\u884c\u8a13\u7df4\uff0c\u8403\u7df4\u9ad8\u968e\u7684\u8a9e\u610f\u60c5\u7dd2\u7279\u5fb5\u3002 (3)\u7531 2D-CNN \u548c BERT \u5b78\u7fd2\u7684\u8072\u5b78\u548c\u8a9e\u610f\u60c5\u7dd2\u7279\u5fb5\u53c3\u6578\uff0c\u88ab\u96c6\u6210\u5728\u591a\u6a21\u5f0f\u7684\u878d\u5408\u7db2\u7d61\u4e2d\u3002\u6700\u5f8c\uff0c \u6211\u5011\u63a1\u7528\u591a\u6a21\u5f0f\u7684\u6700\u5f8c\u4e00\u500b\u96b1\u85cf\u5c64\u7684\u8f38\u51fa\u4f5c\u70ba\u5206\u6bb5\u7684\u60c5\u611f\u6a19\u7c64\u3002 \u4f5c\u7528\u3002\u7531\u5716\u4e09\u4e2d\u53ef\u4ee5\u770b\u51fa\uff0c\u672c\u8ad6\u6587\u5c07 MFCCs \u4f5c\u70ba 2D \u653e\u662f\u4f5c\u7232\u8f38\u5165\uff0c\u8f38\u5165\u7dca\u63a5\u516d\u5c64 CNN \u7684\u57fa\u672c\u5c64\u6578\uff0c\u5982\u5716\u4e09\u6240\u793a\uff0cCNN \u5177\u6709[INPUT-CONV-RELU-POOL-CONV-RELUPOOL] \u7684\u57fa\u672c\u67b6\u69cb\u3002CNN \u8f38\u5165\u7684\u5927\u5c0f\u70ba 40 * 32\uff0c\u70ba\u4e86\u76e1\u53ef\u80fd\u4fdd\u7559 Fearless Steps Challenge \u63d0 \u4f9b\u7684\u4fe1\u606f\uff0c\u6211\u5011\u70ba\u6bcf\u500b\u60c5\u7dd2\u8cc7\u6599\u5229\u7528 Sliding Windows \u7a97\u53e3\u63a1\u6a23\u8a13\u7df4\u6578\u64da\u3002\u6700\u5f8c\uff0c\u5c07\u5b8c \u6574\u7684 CNN \u67b6\u69cb\u97f3\u983b\u8b58\u5225\u7684\u90e8\u5206\u52a0\u5165\u6df7\u5408\u795e\u7d93\u7db2\u8def\u6a21\u578b\u67b6\u69cb\u3002 \u56db\u3001\u60c5\u7dd2\u5075\u6e2c\u5206\u985e\u5be6\u9a57 (\u4e00)\u8a13\u7df4\u8207\u6e2c\u8a66\u8a9e\u6599 \u9577\u6642\u8a13\u7df4\u53ca\u8fa8\u8b58\u7684\u6587\u5b57\u548c\u8a9e\u97f3\u7531 Fearless Steps Challenge \u6240\u63d0\u4f9b\u7684\u7f8e\u570b\u5b87\u822a\u5c40\u963f\u6ce2 \u5716\u4e94\u3001\u5f97\u5206\u7bc4\u570d\u53c3\u8003\u5716 \u8a55\u5206\u516c\u5f0f\u5982\u4e0b\uff0c \u70ba System Detected \u7684\u771f\u5be6\u5f97\u5206\u7684\u7e3d\u6642\u9593\u548c\uff0c \u210e \u70ba Reference annotation \u7684\u53c3\u8003\u7b54\u6848\u6642\u9593\u7e3d\u548c\uff0c \u9664\u4ee5 \u210e \u518d\u4e58\u4ee5\u767e\u5206\u6bd4\u70ba\u6700\u5f8c\uff0cFearless Steps \u5716\u516d\u3001Fearless Steps Challenge \u5b98\u65b9\u6392\u540d\u7e3d\u8868 \u5be6\u9a57\u4e00\uff0c\u8072\u5b78\u8207\u6587\u5b57\u6a21\u5f0f\u60c5\u7dd2\u5075\u6e2c 1. \u8072\u5b78 CNN \u6a21\u578b \u591a\u6a21\u5f0f\u6a21\u578b\u97f3\u983b\u524d\u53ca\u8655\u7406\u90e8\u5206\u55ae\u7368\u9032\u884c\u8a0e\u8ad6\uff0c Fearless Steps Challenge \u7684\u7b54\u6848\u5171\u5206 40.07 41.2 Acc \u8f38\u51fa\u8da8\u8fd1\u65bc\u7a69\u5b9a\u3002 42.05 43.04 43.47 43.24 45.46 44.9 45.22 45.37 45.14 45.04 45.93 44.46 45.32 45.28 45.84 45.92 45.39 45.33 \u6703\u5224\u65b7\u8f38\u51fa\u70ba 1\uff0c\u7576\uff24\u8f49\u614b\u51fa\u73fe\u4e2d\u65b7\u6216\u662f\u5c0f\u65bc 3 \u6b21\u6642\u56de\u9053\u539f\u59cb\u72c0\u614b\u7684 VAD \u503c\uff0c\u53cd\u4e4b\u5247 \u5716\u516b\u3001state machine \u524d\u5f8c\u6bd4\u8f03 45.25 \u5c07\u72c0\u614b\u8f49\u70ba\u8f49\u614b\u6578\u503c\u3002\u4e5f\u5c31\u662f\u8aaa\uff0c\u7576\u8f38\u51fa\u7b2c\u4e8c\u6b21\u51fa\u73fe\u4e0d\u4e00\u6a23\u7684\u6578\u503c\u6642\u5148\u653e\u5165\u66ab\u5b58\u5668\uff0c\u7136 \u5728 state machine \u7684\u5e6b\u52a9\u4e0b\uff0c\u672c\u5be6\u9a57\u4f7f\u7528\u5716\u4e94\u7684\u591a\u6a21\u5f0f\u795e\u7d93\u7db2\u8def\u67b6\u69cb\uff0c\u5c07\u6587\u5b57\u4ee5\u53ca\u8072 43.6 \u800c\u7e7c\u7e8c\u8f38\u51fa\u76f8\u540c\u6578\u503c\uff0c\u76f4\u5230\u9023\u7e8c\u5f97\u5230\u76f8\u540c\u8f49\u614b\u6578\u503c\uff0c\u624d\u78ba\u5b9a\u8f49\u614b\u3002\u9019\u53ef\u4ee5\u4f7f\u672c\u8ad6\u6587\u6a21\u578b \u97f3\u4f7f\u7528</td></tr><tr><td>\u60c5\u7dd2\u7279\u5fb5\uff0c\u53ef\u80fd\u540c\u6642\u8868\u73fe\u5728\u97f3\u8272\u3001\u8a9e\u6c23\u6216\u662f\u6587\u5b57\u7528\u8a9e\u4e0a\u3002\u56e0\u6b64\u5728\u6bd4\u8cfd\u7576\u4e2d\uff0c\u6211\u5011\u9664\u4e86\u5206 \u4e8c\u3001Fearless Steps Challenge \u7f85\u8a08\u5283\u7684\u5168\u7a0b\u7121\u7dda\u96fb\u8cc7\u6599\u5eab\uff0c\u5305\u62ec 100 \u500b\u5c0f\u6642\u3002\u9078\u64c7\u7684\u963f\u6ce2\u7f85 11 \u865f\u4efb\u52d9\u4e3b\u8981\u5206\u70ba\u4e09\u500b Challenge \u6bd4\u8cfd\u6392\u540d\u7684\u53c3\u8003\u4f9d\u64da\u3002 \u70ba\u4e09\u7a2e positive\u3001neutral\u3001negative\uff0c\u800c\u5728\u8a55\u4f30\u6307\u6a19\u5167\u9084\u6709\u5305\u542b Non-Sentiment \u7684\u90e8\u5206\uff0c\u56e0</td></tr><tr><td>\u5225\u5690\u8a66\u5c0d\u65bc\u8072\u97f3\u548c\u9010\u5b57\u7a3f\u62bd\u53d6\u5176\u96b1\u542b\u7684\u60c5\u7dd2\u76f8\u95dc\u7279\u5fb5\uff0c\u4e26\u5e0c\u671b\u4ee5\u591a\u6a21\u5f0f\u795e\u7d93\u7db2\u8def\uff0c\u5c07\u5169 ECOM FD GNC MOCR NTWK \u968e\u6bb5\uff1a (i)\u5347\u7a7a\u3001 (ii)\u767b\u6708\u3001 (iii)\u6708\u7403\u884c\u8d70\u3002\u70ba\u4efb\u52d9\u7cfb\u7d71\u958b\u767c\u63d0\u4f9b\u4e86 80 \u5c0f\u6642\u7684\u97f3\u983b\u3002 \u6b64\u5728\u8a13\u7df4\u540c\u6642\u5c07\u6e2c\u8a66\u96c6 Non-Sentiment \u7684\u90e8\u5206\u4f7f\u7528 Sliding Windows \u9032\u884c\u6578\u64da\u63a1\u96c6\u3002</td></tr><tr><td>\u8005\u7684\u7279\u5fb5\u53c3\u6578\u9032\u884c\u7d50\u5408\uff0c\u540c\u6642\u4ee5\u8072\u97f3\u4e2d\u8207\u9010\u5b57\u7a3f\u4e2d\u7684\u60c5\u7dd2\u7279\u5fb5\u4f86\u5efa\u7acb\u6a21\u578b\uff0c\u4ee5\u63d0\u5347\u60c5\u7dd2 \u5075\u6e2c\u7684\u6b63\u78ba\u7387\u3002 (\u4e00)\u3001\u6578\u64da\u96c6 \u70ba\u4e86\u8a55\u4f30\u672c\u6587\u6240\u63d0\u51fa\u7684\u6a21\u578b\u6027\u80fd\uff0c\u6211\u5011\u4f7f\u7528 Fearless Steps Challenge \u6240\u63d0\u4f9b\u7684\u7f8e\u570b SNR (Mean) 13.32 14.67 14.91 5.07 10.68 SNR (Std. Dev) 7.40 10.51 11.96 12.60 \u5728\u9019 80 \u500b\u5c0f\u6642\u5167\uff0c\u63d0\u4f9b\u4e86 20 \u5c0f\u6642\u7684\u7d93\u904e\u4eba\u5de5\u9a57\u8b49\u7684\u7b54\u6848\u3002\u5c0d\u65bc\u5269\u9918\u7684 60 \u5c0f\u6642\u97f3\u983b\uff0c \uff21 \uff1d \u210e \u5728\u97f3\u983b\u6e2c\u8a66\u4e2d\u53ef\u4ee5\u770b\u5230\uff0c\u56e0\u8cc7\u6599\u5eab\u97f3\u6a94\u96dc\u8a0a\u904e\u591a\u4e14\u5728\u5927\u90e8\u5206\u97f3\u6a94\u7576\u4e2d\u7684\u5c0d\u8a71\u60c5\u7dd2\u8d77 11.17 \u63d0\u4f9b Baseline \u7cfb\u7d71\u751f\u6210\u7684\u8f38\u51fa\u7b54\u6848\uff0c\u53e6\u5916\u4e00\u7d44 20 \u5c0f\u6642\u5c07\u767c\u5e03\u7528\u65bc\u958b\u653e\u6e2c\u8a66\u3002 \u4f0f\u4e26\u4e0d\u660e\u986f\uff0c\u6240\u4ee5\u9020\u6210 positive\u3001negative \u7684\u6e96\u78ba\u7387\u504f\u4f4e\uff0c\u4f46\u5728 neutral\u3001silence \u7684\u90e8\u5206\u4ee5 Txt Range</td></tr><tr><td>\u5b87\u822a\u5c40\u963f\u6ce2\u7f85\u8a08\u5283\u7684\u5168\u7a0b\u7121\u7dda\u96fb\u901a\u8a0a\u9304\u97f3\u8cc7\u6599\u5eab\uff0c\u5171\u6709 100 \u500b\u5c0f\u6642\uff0c\u5305\u62ec\u706b\u7bad\u5347\u7a7a\u7d04\u4f54 \u5716\u4e03\u6df7\u6dc6\u77e9\u9663\u4f86\u770b silence \u7684\u6e96\u78ba\u7387\u6700\u9ad8\uff0c\u56e0\u6b64\u5728\u55ae\u97f3\u983b\u6e2c\u8a66\u6a21\u578b\u4e0b\u6210\u6548\u8f03\u70ba\u986f\u8457\uff0c\u4f46</td></tr></table>" | |
| } | |
| } | |
| } | |
| } |