{ "paper_id": "O13-1016", "header": { "generated_with": "S2ORC 1.0.0", "date_generated": "2023-01-19T08:03:21.877707Z" }, "title": "Text-independent Speaker Verification using a Hybrid I-Vector/DNN Approach \u7d50\u5408 I-Vector \u53ca\u6df1\u5c64\u795e\u7d93\u7db2\u8def\u4e4b\u8a9e\u8005\u9a57\u8b49\u7cfb\u7d71", "authors": [ { "first": "Yun-Fan", "middle": [], "last": "\u5f35\u96f2\u5e06", "suffix": "", "affiliation": {}, "email": "" }, { "first": "", "middle": [], "last": "Chang", "suffix": "", "affiliation": {}, "email": "wtchang@iii.org.tw" }, { "first": "Yu", "middle": [], "last": "\uff0c\u66f9\u6631", "suffix": "", "affiliation": {}, "email": "" }, { "first": "", "middle": [], "last": "Tsao", "suffix": "", "affiliation": {}, "email": "" }, { "first": "Shao-Hua", "middle": [], "last": "\uff0c\u912d\u5c11\u6a3a", "suffix": "", "affiliation": {}, "email": "" }, { "first": "", "middle": [], "last": "Cheng", "suffix": "", "affiliation": {}, "email": "" }, { "first": "Kai-Hsuan", "middle": [], "last": "\uff0c\u8a79\u51f1\u8ed2", "suffix": "", "affiliation": {}, "email": "" }, { "first": "", "middle": [], "last": "Chan", "suffix": "", "affiliation": {}, "email": "" }, { "first": "Chia-Wei", "middle": [], "last": "\uff0c\u5ed6\u5609\u7dad", "suffix": "", "affiliation": {}, "email": "" }, { "first": "", "middle": [], "last": "Liao", "suffix": "", "affiliation": {}, "email": "" }, { "first": "Wen-Tsung", "middle": [], "last": "\uff0c\u5f35\u6587\u6751", "suffix": "", "affiliation": {}, "email": "" }, { "first": "", "middle": [], "last": "\u8ca1\u5718\u6cd5\u4eba\u8cc7\u8a0a\u5de5\u696d\u7b56\u9032\u6703\u524d\u77bb\u79d1\u6280\u7814\u7a76\u6240", "suffix": "", "affiliation": {}, "email": "" } ], "year": "", "venue": null, "identifiers": {}, "abstract": "", "pdf_parse": { "paper_id": "O13-1016", "_pdf_hash": "", "abstract": [], "body_text": [], "back_matter": [], "bib_entries": { "BIBREF0": { "ref_id": "b0", "title": "Text-independent speaker identification", "authors": [ { "first": "H", "middle": [], "last": "Gish", "suffix": "" }, { "first": "M", "middle": [], "last": "Schmidt", "suffix": "" } ], "year": 1994, "venue": "IEEE, Signal Processing Magazine", "volume": "11", "issue": "", "pages": "18--32", "other_ids": {}, "num": null, "urls": [], "raw_text": "H. Gish and M. Schmidt, \"Text-independent speaker identification,\" IEEE, Signal Processing Magazine, vol. 11, pp. 18-32, 1994.", "links": null }, "BIBREF1": { "ref_id": "b1", "title": "Speaker verification using adapted Gaussian mixture models", "authors": [ { "first": "D", "middle": [ "A" ], "last": "Reynolds", "suffix": "" }, { "first": "T", "middle": [ "F" ], "last": "Quatieri", "suffix": "" }, { "first": "R", "middle": [ "B" ], "last": "Dunn", "suffix": "" } ], "year": 2000, "venue": "Digital Signal Processing", "volume": "10", "issue": "", "pages": "19--41", "other_ids": {}, "num": null, "urls": [], "raw_text": "D. A. Reynolds, T. F. Quatieri, and R. B. Dunn, \"Speaker verification using adapted Gaussian mixture models,\" ELSEVIER, Digital Signal Processing, vol. 10, pp. 19-41, 2000.", "links": null }, "BIBREF2": { "ref_id": "b2", "title": "An overview of automatic speaker recognition technology", "authors": [ { "first": "D", "middle": [ "A" ], "last": "Reynolds", "suffix": "" } ], "year": 2002, "venue": "IEEE Transactions, Acoustics Speech and Signal Processing", "volume": "4", "issue": "", "pages": "4072--4075", "other_ids": {}, "num": null, "urls": [], "raw_text": "D. A. Reynolds, \"An overview of automatic speaker recognition technology,\" IEEE Transactions, Acoustics Speech and Signal Processing, vol. 4, pp. 4072-4075, 2002.", "links": null }, "BIBREF3": { "ref_id": "b3", "title": "A hybrid GMM/SVM approach to speaker identification", "authors": [ { "first": "S", "middle": [], "last": "Fine", "suffix": "" }, { "first": "J", "middle": [], "last": "Navratil", "suffix": "" }, { "first": "R", "middle": [ "A" ], "last": "Gopinath", "suffix": "" } ], "year": 2001, "venue": "IEEE Transactions, Acoustics Speech and Signal Processing", "volume": "1", "issue": "", "pages": "417--420", "other_ids": {}, "num": null, "urls": [], "raw_text": "S. Fine, J. Navratil and R. A. Gopinath, \"A hybrid GMM/SVM approach to speaker identification,\" IEEE Transactions, Acoustics Speech and Signal Processing, vol. 1, pp. 417-420, 2001.", "links": null }, "BIBREF4": { "ref_id": "b4", "title": "Support vector machines for speaker and language recognition", "authors": [ { "first": "W", "middle": [ "M" ], "last": "Campbell", "suffix": "" }, { "first": "J", "middle": [ "P" ], "last": "Campbell", "suffix": "" }, { "first": "D", "middle": [ "A" ], "last": "Reynolds", "suffix": "" }, { "first": "E", "middle": [], "last": "Singer", "suffix": "" }, { "first": "P", "middle": [ "A" ], "last": "Torres-Carrasquillo", "suffix": "" } ], "year": 2006, "venue": "Computer Speech & Language", "volume": "20", "issue": "", "pages": "210--229", "other_ids": {}, "num": null, "urls": [], "raw_text": "W. M. Campbell, J. P. Campbell, D. A. Reynolds, E. Singer, and P. A. Torres-Carrasquillo, \"Support vector machines for speaker and language recognition,\" ELSEVIER, Computer Speech & Language, vol. 20, pp. 210-229, 2006.", "links": null }, "BIBREF5": { "ref_id": "b5", "title": "Front-end factor analysis for speaker verification", "authors": [ { "first": "N", "middle": [], "last": "Dehak", "suffix": "" }, { "first": "P", "middle": [], "last": "Kenny", "suffix": "" }, { "first": "R", "middle": [], "last": "Dehak", "suffix": "" }, { "first": "P", "middle": [], "last": "Dumouchel", "suffix": "" }, { "first": "P", "middle": [], "last": "Ouellet", "suffix": "" } ], "year": 2011, "venue": "speech ,and Language Processing", "volume": "19", "issue": "", "pages": "788--798", "other_ids": {}, "num": null, "urls": [], "raw_text": "N. Dehak, P. Kenny, R. Dehak, P. Dumouchel, and P. Ouellet, \"Front-end factor analysis for speaker verification,\" IEEE Transactions, Audio, speech ,and Language Processing, vol. 19, pp. 788-798, 2011.", "links": null }, "BIBREF6": { "ref_id": "b6", "title": "Exploring strategies for training deep neural networks", "authors": [ { "first": "H", "middle": [], "last": "Larochelle", "suffix": "" }, { "first": "Y", "middle": [], "last": "Benqio", "suffix": "" }, { "first": "J", "middle": [], "last": "Louradour", "suffix": "" }, { "first": "P", "middle": [], "last": "Lamblin", "suffix": "" } ], "year": 2009, "venue": "Machine Learning", "volume": "10", "issue": "", "pages": "1--40", "other_ids": {}, "num": null, "urls": [], "raw_text": "H. Larochelle, Y. Benqio, J. Louradour, and P. Lamblin, \"Exploring strategies for training deep neural networks,\" Machine Learning, vol. 10, pp. 1-40, 2009.", "links": null }, "BIBREF7": { "ref_id": "b7", "title": "Deep neural networks for acoustic modeling in speech recognition: the shared views of four research groups", "authors": [ { "first": "G", "middle": [], "last": "Hinton", "suffix": "" }, { "first": "L", "middle": [], "last": "Deng", "suffix": "" }, { "first": "D", "middle": [], "last": "Yu", "suffix": "" }, { "first": "G", "middle": [ "E" ], "last": "Dahl", "suffix": "" }, { "first": "A", "middle": [], "last": "Mohamed", "suffix": "" }, { "first": "N", "middle": [], "last": "Jaitly", "suffix": "" }, { "first": "A", "middle": [], "last": "Senior", "suffix": "" }, { "first": "V", "middle": [], "last": "Vanhoucke", "suffix": "" }, { "first": "P", "middle": [], "last": "Nguyen", "suffix": "" }, { "first": "T", "middle": [ "N" ], "last": "Sainath", "suffix": "" }, { "first": "B", "middle": [], "last": "Kingsbury", "suffix": "" } ], "year": 2012, "venue": "IEEE, Signal Processing Magazine", "volume": "29", "issue": "", "pages": "82--97", "other_ids": {}, "num": null, "urls": [], "raw_text": "G. Hinton, L. Deng, D. Yu, G. E. Dahl, A. Mohamed, N. Jaitly, A. Senior, V. Vanhoucke, P. Nguyen, T. N. Sainath, and B. Kingsbury, \"Deep neural networks for acoustic modeling in speech recognition: the shared views of four research groups,\" IEEE, Signal Processing Magazine, vol. 29, pp. 82-97, 2012.", "links": null }, "BIBREF8": { "ref_id": "b8", "title": "Learning deep architectures for AI", "authors": [ { "first": "Y", "middle": [], "last": "Bengio", "suffix": "" } ], "year": 2009, "venue": "Found. Trends Mach. Learn", "volume": "2", "issue": "", "pages": "1--127", "other_ids": {}, "num": null, "urls": [], "raw_text": "Y. Bengio, \"Learning deep architectures for AI,\" Found. Trends Mach. Learn, vol. 2, pp. 1-127, 2009.", "links": null }, "BIBREF9": { "ref_id": "b9", "title": "Acoustic modeling using deep belief networks", "authors": [ { "first": "A", "middle": [], "last": "Mohamed", "suffix": "" }, { "first": "G", "middle": [ "E" ], "last": "Dahl", "suffix": "" }, { "first": "G", "middle": [ "E" ], "last": "Hinton", "suffix": "" } ], "year": 2013, "venue": "speech ,and Language Processing", "volume": "20", "issue": "", "pages": "14--22", "other_ids": {}, "num": null, "urls": [], "raw_text": "A. Mohamed, G. E. Dahl, and G. E. Hinton, \"Acoustic modeling using deep belief networks,\" IEEE Transactions, Audio, speech ,and Language Processing, 20, 14-22, 2013.", "links": null } }, "ref_entries": { "TABREF0": { "content": "
\u6458\u8981 \u8a9e\u8005\u9a57\u8b49\u7684\u76ee\u7684\u662f\u4ee5\u8a9e\u97f3\u8a0a\u865f\u4f86\u9a57\u8b49\u7279\u5b9a\u8a9e\u8005\u7684\u8eab\u4efd(Identity)\uff0c\u6b64\u9805\u7814\u7a76\u5728\u8fd1\u5e74\u7684 \u667a\u6167\u751f\u6d3b\u74b0\u5883\u5df2\u6210\u70ba\u4e00\u500b\u91cd\u8981\u7684\u7814\u7a76\u8b70\u984c\u3002\u4e0d\u8ad6\u662f\u5728\u9580\u7981\u7cfb\u7d71\uff0c\u4ea6\u6216\u662f\u641c\u5c0b\u3001\u5075\u6e2c\u7279\u5b9a \u8a9e\u8005\u8a9e\u97f3\u7b49\uff0c\u90fd\u88ab\u5ee3\u6cdb\u61c9\u7528\u3002\u8a9e\u8005\u9a57\u8b49\u53c8\u5206\u70ba\u6587\u5b57\u7279\u5b9a\u6a21\u5f0f(Test-dependent Mode)\u8207\u6587 \u5b57\u4e0d\u7279\u5b9a\u6a21\u5f0f(Text-independent Mode) \u5169\u985e [1]\uff0c\u524d\u8005\u7684\u597d\u8655\u70ba\u5df2\u77e5\u8f03\u591a\u8a9e\u97f3\u8cc7\u8a0a\uff0c\u53ef \u4ee5\u5927\u5e45\u6539\u5584\u7cfb\u7d71\u7684\u9a57\u8b49\u6548\u80fd\uff0c\u4f46\u5be6\u969b\u7684\u61c9\u7528\u9650\u5236\u8f03\u591a\uff0c\u5f8c\u8005\u56e0\u70ba\u662f\u96a8\u6a5f\u7684\u8a9e\u97f3\u8a0a\u865f\uff0c\u8cc7 \u8a0a\u91cf\u8f03\u5c11\uff0c\u76f8\u5c0d\u9a57\u8b49\u6548\u679c\u4e0d\u5982\u524d\u8005\uff0c\u4f46\u4e5f\u56e0\u70ba\u9650\u5236\u8f03\u5c11\uff0c\u61c9\u7528\u5c64\u9762\u76f8\u5c0d\u8f03\u5927\u3002\u5728\u672c\u7814\u7a76 \u4e2d\uff0c\u6211\u5011\u8457\u91cd\u65bc\u6587\u5b57\u4e0d\u7279\u5b9a\u6a21\u5f0f\u7684\u8a9e\u8005\u9a57\u8b49\u3002 \u50b3\u7d71\u7684\u8a9e\u8005\u9a57\u8b49\u7cfb\u7d71\u662f\u4f7f\u7528\u9ad8\u65af\u6df7\u5408\u6a21\u578b\u7684\u67b6\u69cb\uff0c\u5176\u4f5c\u6cd5\u662f\u8a13\u7df4\u4e00\u5957 Universal Background Model (UBM)\u9ad8\u65af\u6df7\u5408\u6a21\u578b(Gaussian Mixture Model\uff0cGMM)\uff0cUBM-GMM\u3002 \u63a5\u8457\u5229\u7528\u6bcf\u4e00\u4f4d\u8a9e\u8005\u7684\u8a9e\u97f3\u8a0a\u865f\uff0c\u4ee5\u53ca\u6700\u5927\u5f8c\u9a57\u6982\u7387\u6cd5\u5247(Maximum A Posteriori\uff0cMAP) Recall F-measure Accuracy EER SVM 35% 67% 46% 84% 19.26% DNN 70% 70% 70% 94% 12.22% \u5c0d Precision \u53c3\u8003\u6587\u737b
", "type_str": "table", "num": null, "html": null, "text": "UBM-GMM \u4f5c\u8abf\u6574\u4ee5\u5f97\u5230\u6bcf\u4f4d\u8a9e\u8005\u5c08\u5c6c\u6a21\u578b\uff0c\u63a5\u8457\u518d\u5c0d\u6e2c\u8a66\u8a9e\u53e5\u5229\u7528 UBM-GMM \u53ca Speaker-specific GMM \u5206\u5225\u8a08\u7b97\u4f3c\u7136\u503c [2, 3]\u3002\u53e6\u5916\uff0c\u9084\u6709\u5c07 GMM \u62bd\u53d6 Mean \u4e32\u6210 Supervector \u518d\u4f7f\u7528 Support Vector Machine(SVM)\u4f5c\u8fa8\u8b58\u7684\u65b9\u6cd5 [4, 5]\u3002 \u8fd1\u5e74\u4f86\u5728 NIST Speaker Recognition Evaluations(SRE)\u767c\u5c55\u4e86\u4e00\u5957 I-Vector \u7684\u7279\u5fb5\u64f7 \u53d6\u65b9\u5f0f\uff0c\u5176\u7279\u5fb5\u64f7\u53d6\u5305\u542b\u4ee5\u4e0b\u4e09\u500b\u6b65\u9a5f:1.\u5c0d\u8a9e\u97f3\u8a0a\u865f\u4f5c MFCC \u7279\u5fb5\u64f7\u53d6\u30022.\u5229\u7528 UBM-GMM \u8a08\u7b97\u51fa\u6bcf\u4f4d\u8a9e\u8005\u7684 Supervector\u30023.\u4f7f\u7528 Baum-Welch Statistics \u8a08\u7b97\u51fa I-Vector\u3002\u904e\u53bb\u7684\u7814\u7a76\u8b49\u5be6\uff0cI-Vector \u642d\u914d SVM \u5206\u985e\u5668\uff0c\u80fd\u6709\u6548\u5730\u5b8c\u6210\u8a9e\u8005\u8b58\u5225 [6]\u3002 \u8fd1\u65e5\uff0c\u6df1\u5c64\u795e\u7d93\u7db2\u8def(Deep Neural Network\uff0cDNN)\u5df2\u88ab\u5ee3\u6cdb\u5730\u61c9\u7528\u5728\u5404\u985e\u578b\u7684\u5206\u985e\u554f\u984c [7-10]\u3002\u672c\u8ad6\u6587\u63d0\u51fa\u4f7f\u7528 I-Vector \u7d50\u5408\u6df1\u5c64\u795e\u7d93\u7db2\u8def\u9032\u884c\u8a9e\u8005\u9a57\u8b49\u3002 \u672c\u5be6\u9a57\u6240\u4f7f\u7528\u7684\u8cc7\u6599\u70ba\u67d0\u8ac7\u8a71\u6027\u7bc0\u76ee\u5be6\u969b\u8a9e\u97f3\u8cc7\u6599\uff0c\u76ee\u7684\u70ba\u627e\u51fa\u7279\u5b9a\u5973\u4e3b\u6301\u4eba\u7684\u8a9e \u97f3\u7247\u6bb5\u3002\u8a13\u7df4\u8a9e\u6599\u70ba 177 \u53e5\u5973\u6027\u8ac7\u8a71\u8a9e\u53e5\uff0c\u76ee\u6a19\u8a13\u7df4\u8a9e\u53e5\u70ba\u67d0\u5973\u4e3b\u6301\u4eba\u7684 12 \u53e5\u8a9e\u6599\uff0c \u5176\u9577\u5ea6\u5747\u7d04 6 \u79d2\u3002\u7d93\u904e Voice Activity Detection(VAD)\u8655\u7406\u5f8c\uff0c\u8a13\u7df4\u8a9e\u6599\u5207\u5272\u6210 1,921 \u53e5 \u8a9e\u53e5\uff0c\u76ee\u6a19\u8a13\u7df4\u8a9e\u53e5\u5207\u5272\u6210 118 \u53e5\u8a9e\u53e5\u3002\u6e2c\u8a66\u8a9e\u6599\u5171 300 \u53e5\uff0c\u5176\u4e2d 30 \u53e5\u70ba\u76ee\u6a19\u8a9e\u53e5\uff0c \u5176\u9918 270 \u53e5\u70ba\u7537\u5973\u6df7\u5408\u4e4b\u8a9e\u6599\uff0c\u9577\u5ea6\u5747\u7d04 3 \u79d2\u3002\u5be6\u9a57\u8a2d\u5b9a\u90e8\u5206\uff0cMFCC \u7279\u5fb5\u70ba 13 \u7dad\u5c55 \u5ef6\u6210 39 \u7dad MFCC\uff0cI-Vector \u4f7f\u7528 256 \u500b\u9ad8\u65af\u6df7\u5408\u6578\u7684 UBM-GMM\uff0c\u5176\u7dad\u5ea6\u70ba 64 \u7dad\u3002\u5728 \u6b64\u7bc7\u8ad6\u6587\u7684\u5be6\u9a57\u7d50\u679c\u986f\u793a\uff0c\u589e\u52a0\u7dad\u5ea6\u4e0d\u6703\u660e\u986f\u63d0\u5347\u8fa8\u8b58\u7d50\u679c\uff0c\u800c\u76f8\u5c0d\u6703\u7522\u751f\u984d\u5916\u7684\u904b\u7b97 \u91cf\u3002DNN \u4f7f\u7528\u5169\u5c64\u96b1\u85cf\u5c64\uff0c\u795e\u7d93\u55ae\u5143\u5747\u8a2d\u70ba 150\uff0c\u5176\u539f\u56e0\u8207\u4e0a\u8ff0\u76f8\u540c\u3002 \u5728\u5be6\u9a57\u4e2d\uff0c\u5c0d\u65bc\u8a9e\u8005\u9a57\u8b49\u7cfb\u7d71\u7684\u8a55\u91cf\uff0c\u6211\u5011\u901a\u5e38\u4f7f\u7528 Equal Error Rate(EER)\u505a\u70ba\u8a55 \u91cf\u6a19\u6e96\u3002\u53e6\u5916\uff0c\u6211\u5011\u9084\u4f7f\u7528 Precision\u3001Recall\u3001F-measure \u548c Accuracy \u8a55\u91cf\u6a21\u578b\u6548\u80fd\uff0c \u6211\u5011\u5c07\u5be6\u9a57\u7d50\u679c\u6574\u7406\u65bc\u4e0b\u8868\u4e00\u3002\u7531\u5be6\u9a57\u7d50\u679c\u53ef\u77e5\uff0c\u6211\u5011\u63d0\u51fa\u7684 I-Vector \u642d\u914d DNN \u7cfb\u7d71 \u5728\u5404\u7a2e\u8a55\u91cf\u65b9\u5f0f\u7686\u512a\u65bc I-Vector \u642d\u914d SVM \u7cfb\u7d71\u3002 \u8868\u4e00\u3001\u8a55\u4f30\u7d50\u679c" } } } }