| { | |
| "paper_id": "O16-3004", | |
| "header": { | |
| "generated_with": "S2ORC 1.0.0", | |
| "date_generated": "2023-01-19T08:04:48.902675Z" | |
| }, | |
| "title": "Evaluation Metric-related Optimization Methods for Mandarin Mispronunciation Detection", | |
| "authors": [ | |
| { | |
| "first": "\u8a31\u66dc\u9e92", | |
| "middle": [ | |
| "\uf02a" | |
| ], | |
| "last": "\u3001\u694a\u660e\u7ff0", | |
| "suffix": "", | |
| "affiliation": {}, | |
| "email": "" | |
| }, | |
| { | |
| "first": "\uf02a", | |
| "middle": [], | |
| "last": "\u3001\u6d2a\u5b5d\u5b97", | |
| "suffix": "", | |
| "affiliation": {}, | |
| "email": "" | |
| }, | |
| { | |
| "first": "\uf02a", | |
| "middle": [], | |
| "last": "\u3001\u6797\u5955\u5112", | |
| "suffix": "", | |
| "affiliation": {}, | |
| "email": "" | |
| }, | |
| { | |
| "first": "\uf02a", | |
| "middle": [], | |
| "last": "\u3001\u9673\u51a0\u5b87", | |
| "suffix": "", | |
| "affiliation": {}, | |
| "email": "" | |
| }, | |
| { | |
| "first": "\uf02b", | |
| "middle": [], | |
| "last": "\u3001\u9673\u67cf\u7433", | |
| "suffix": "", | |
| "affiliation": {}, | |
| "email": "" | |
| }, | |
| { | |
| "first": "Yao-Chi", | |
| "middle": [], | |
| "last": "Hsu", | |
| "suffix": "", | |
| "affiliation": {}, | |
| "email": "[email protected]" | |
| }, | |
| { | |
| "first": "Ming-Han", | |
| "middle": [], | |
| "last": "Yang", | |
| "suffix": "", | |
| "affiliation": {}, | |
| "email": "[email protected]" | |
| }, | |
| { | |
| "first": "Hsiao-Tsung", | |
| "middle": [], | |
| "last": "Hung", | |
| "suffix": "", | |
| "affiliation": {}, | |
| "email": "" | |
| }, | |
| { | |
| "first": "Yi-Ju", | |
| "middle": [], | |
| "last": "Lin", | |
| "suffix": "", | |
| "affiliation": {}, | |
| "email": "[email protected]" | |
| }, | |
| { | |
| "first": "Kuan-Yu", | |
| "middle": [], | |
| "last": "Chen", | |
| "suffix": "", | |
| "affiliation": {}, | |
| "email": "[email protected]" | |
| }, | |
| { | |
| "first": "Berlin", | |
| "middle": [], | |
| "last": "Chen", | |
| "suffix": "", | |
| "affiliation": {}, | |
| "email": "[email protected]" | |
| } | |
| ], | |
| "year": "", | |
| "venue": null, | |
| "identifiers": {}, | |
| "abstract": "Mispronunciation detection and diagnosis are part and parcel of a computer assisted pronunciation training (CAPT) system, collectively facilitating second-language (L2) learners to pinpoint erroneous pronunciations in a given utterance so as to improve their spoken proficiency. This thesis presents a continuation of such a general line of research and the major contributions are threefold. First, we compared the performance of different pronunciation features in mispronunciation detection. Second, we propose an effective training approach that estimates the deep neural network based acoustic models involved in the mispronunciation detection process by optimizing an objective directly linked to the ultimate evaluation metric. Third, we can linearly combine two F 1-score when we consider F 1-score as final objective function. It can effectively deal with the label imbalance problem. A series of experiments on a Mandarin mispronunciation detection task seem to show the performance merits of the proposed methods.", | |
| "pdf_parse": { | |
| "paper_id": "O16-3004", | |
| "_pdf_hash": "", | |
| "abstract": [ | |
| { | |
| "text": "Mispronunciation detection and diagnosis are part and parcel of a computer assisted pronunciation training (CAPT) system, collectively facilitating second-language (L2) learners to pinpoint erroneous pronunciations in a given utterance so as to improve their spoken proficiency. This thesis presents a continuation of such a general line of research and the major contributions are threefold. First, we compared the performance of different pronunciation features in mispronunciation detection. Second, we propose an effective training approach that estimates the deep neural network based acoustic models involved in the mispronunciation detection process by optimizing an objective directly linked to the ultimate evaluation metric. Third, we can linearly combine two F 1-score when we consider F 1-score as final objective function. It can effectively deal with the label imbalance problem. A series of experiments on a Mandarin mispronunciation detection task seem to show the performance merits of the proposed methods.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Abstract", | |
| "sec_num": null | |
| } | |
| ], | |
| "body_text": [ | |
| { | |
| "text": "\u3002 \u4e0a \u8ff0 \u7684 \u56db \u7a2e \u6307 \u6a19 \u53ef \u4ee5 \u8a08 \u7b97 \u51fa \u5176 \u5b83 \u8a55 \u4f30 \u7684 \u6a19 \u6e96 \uff0c \u4f8b \u5982 \u53ec \u56de \u7387 (Recall) \u8207 \u7cbe \u6e96 \u5ea6 (Precision)\uff0c\u6709\u8a31\u591a\u767c\u97f3\u6aa2\u6e2c\u7684\u7814\u7a76\u7686\u4ee5\u8a72\u8a55\u4f30\u65b9\u5f0f\u4f5c\u70ba\u8a55\u91cf\u7cfb\u7d71\u512a\u52a3\u7684\u6e96\u5247", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": ")\u3002\u6211\u5011\u53ef\u66f4\u9032\u4e00\u6b65\u4f7f\u7528\u53ec\u56de\u7387\u8207\u7cbe\u6e96\u5ea6\u7684\u8abf\u548c\u5e73 \u5747-F \u5ea6\u91cf(F -Score)\u505a\u70ba\u6e96\u5247\uff0cF \u5ea6\u91cf\u5728\u81ea\u7136\u8a9e\u8a00\u8655\u7406(Natural Language Processing, NLP)\u8207\u8cc7\u8a0a\u6aa2\u7d22(Information Retrieval, IR)\u7b49\u7814\u7a76\u4e2d\u5ee3\u70ba\u4f7f\u7528\uff0c\u751a\u81f3\u6709\u8a31\u591a\u4efb\u52d9\u76f4\u63a5\u5c07\u8a72 \u6307\u6a19\u4f5c\u70ba\u6a21\u578b\u8a13\u7df4\u7684\u76ee\u6a19 (Fujino, Isozaki & Suzuki, 2008; Dembczynski, Waegeman, Cheng & H\u00fcllerme\u00eder, 2011; Ye, Chai, Lee & Chieu, 2012 )\u3002\u5728\u932f\u8aa4\u767c\u97f3\u6aa2\u6e2c\u4efb\u52d9\u4e2d\u4e5f\u6709\u985e\u4f3c\u60f3\u6cd5 \u7684 \u7814 \u7a76 \u5df2 \u88ab \u63a2 \u8a0e (Huang et al., 2015; Qian, Soong & Meng, 2010; Huang, Wang & Abudureyimu, 2012) (Hinton et al., 2012) \u3002\u5728\u932f\u8aa4\u767c\u97f3\u6aa2\u6e2c\u7684\u76f8\u95dc\u7814\u7a76\u4e2d\u4e5f\u56e0\u70ba\u6df1\u5c64\u985e\u795e\u7d93\u7db2\u8def \u8072\u5b78\u6a21\u578b\u7684\u4f7f\u7528\u800c\u5728\u6548\u80fd\u4e0a\u6709\u986f\u8457\u7684\u63d0\u5347 (Hu et al., 2015; Qian, Meng & Soong, 2012; Hu et al., 2014) \u3002\u57fa\u65bc\u4e0a\u8ff0\u7814\u7a76\u7684\u555f\u767c\uff0c\u6211\u5011\u5ef6\u7e8c\u904e\u53bb\u5b78\u8005\u4ee5\u6700\u5927\u5316\u932f\u8aa4\u767c\u97f3\u6aa2\u6e2c\u4efb\u52d9\u7684\u6548 \u80fd (Huang et al., 2015; Hsu, Yang, Hung & Chen, 2016) (Kim, Franco & Neumeyer, 1997 (Strik, Truong, De Wet & Cucchiarini, 2007) \u3002\u4f46\u9019\u985e\u7684\u505a\u6cd5\u90fd\u53ea\u6709\u5728\u7279\u5b9a\u7684\u767c\u97f3 \u624d\u80fd\u4f7f\u7528(\u4f8b\u5982\u8377\u862d\u8a9e\u7684/x/\u6216/k/) \uff0c\u4e14\u97fb\u5f8b\u7279\u5fb5\u5bb9\u6613\u56e0\u70ba\u4e0d\u540c\u8a9e\u8005\u800c\u7522\u751f\u7121\u6cd5\u9810\u671f\u7684\u8b8a\u5316\u3002 \u7136\u800c\uff0c\u4e5f\u6709\u5b78\u8005\u57fa\u65bc\u8a9e\u97f3\u8fa8\u8b58\u6a21\u7d44\u4f86\u9032\u884c\u767c\u97f3\u8a3a\u65b7 (Hu et al., 2015) \uff0c\u4f46\u5f9e\u5be6\u9a57\u6578\u64da\u770b\u4f86 \u8ddd\u96e2\u7406\u60f3\u7684\u6e96\u78ba\u7387\u9084\u6709\u4e00\u6bb5\u5dee\u8ddd\u3002\u6709\u4e9b\u5b78\u8005\u8a8d\u70ba\u932f\u8aa4\u767c\u97f3\u6aa2\u6e2c\u8207\u8a3a\u65b7\u61c9\u8a72\u8981\u8996\u70ba\u8a9e\u97f3\u8fa8 \u8b58\u7684\u4efb\u52d9 (Harrison et al., 2008; Harrison et al., 2009; Qian et al., 2012) (Kim et al., 1997) ", | |
| "cite_spans": [ | |
| { | |
| "start": 154, | |
| "end": 186, | |
| "text": "(Fujino, Isozaki & Suzuki, 2008;", | |
| "ref_id": "BIBREF1" | |
| }, | |
| { | |
| "start": 187, | |
| "end": 236, | |
| "text": "Dembczynski, Waegeman, Cheng & H\u00fcllerme\u00eder, 2011;", | |
| "ref_id": null | |
| }, | |
| { | |
| "start": 237, | |
| "end": 264, | |
| "text": "Ye, Chai, Lee & Chieu, 2012", | |
| "ref_id": null | |
| }, | |
| { | |
| "start": 298, | |
| "end": 318, | |
| "text": "(Huang et al., 2015;", | |
| "ref_id": "BIBREF10" | |
| }, | |
| { | |
| "start": 319, | |
| "end": 344, | |
| "text": "Qian, Soong & Meng, 2010;", | |
| "ref_id": "BIBREF18" | |
| }, | |
| { | |
| "start": 345, | |
| "end": 377, | |
| "text": "Huang, Wang & Abudureyimu, 2012)", | |
| "ref_id": "BIBREF9" | |
| }, | |
| { | |
| "start": 378, | |
| "end": 399, | |
| "text": "(Hinton et al., 2012)", | |
| "ref_id": "BIBREF4" | |
| }, | |
| { | |
| "start": 444, | |
| "end": 461, | |
| "text": "(Hu et al., 2015;", | |
| "ref_id": "BIBREF8" | |
| }, | |
| { | |
| "start": 462, | |
| "end": 487, | |
| "text": "Qian, Meng & Soong, 2012;", | |
| "ref_id": "BIBREF17" | |
| }, | |
| { | |
| "start": 488, | |
| "end": 504, | |
| "text": "Hu et al., 2014)", | |
| "ref_id": "BIBREF7" | |
| }, | |
| { | |
| "start": 541, | |
| "end": 561, | |
| "text": "(Huang et al., 2015;", | |
| "ref_id": "BIBREF10" | |
| }, | |
| { | |
| "start": 562, | |
| "end": 591, | |
| "text": "Hsu, Yang, Hung & Chen, 2016)", | |
| "ref_id": "BIBREF6" | |
| }, | |
| { | |
| "start": 592, | |
| "end": 621, | |
| "text": "(Kim, Franco & Neumeyer, 1997", | |
| "ref_id": "BIBREF11" | |
| }, | |
| { | |
| "start": 622, | |
| "end": 665, | |
| "text": "(Strik, Truong, De Wet & Cucchiarini, 2007)", | |
| "ref_id": "BIBREF19" | |
| }, | |
| { | |
| "start": 752, | |
| "end": 769, | |
| "text": "(Hu et al., 2015)", | |
| "ref_id": "BIBREF8" | |
| }, | |
| { | |
| "start": 824, | |
| "end": 847, | |
| "text": "(Harrison et al., 2008;", | |
| "ref_id": "BIBREF2" | |
| }, | |
| { | |
| "start": 848, | |
| "end": 870, | |
| "text": "Harrison et al., 2009;", | |
| "ref_id": "BIBREF3" | |
| }, | |
| { | |
| "start": 871, | |
| "end": 889, | |
| "text": "Qian et al., 2012)", | |
| "ref_id": "BIBREF17" | |
| }, | |
| { | |
| "start": 890, | |
| "end": 908, | |
| "text": "(Kim et al., 1997)", | |
| "ref_id": "BIBREF11" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "\u3002 \u8fd1\u5e74\u4f86\uff0c\u5728\u8a9e\u97f3\u8fa8\u8b58\u7cfb\u7d71\u4e2d\u7684\u8072\u5b78\u6a21\u578b\u5df2\u7531\u6df1\u5c64\u985e\u795e\u7d93\u7db2\u8def(Deep Neural Network, DNN)\u53d6\u4ee3\u50b3\u7d71\u7684\u9ad8\u65af\u6df7\u5408\u6a21\u578b(Gaussian Mixture Model, GMM)\uff0c\u4e26\u5728\u8a9e\u97f3\u8fa8\u8b58\u4efb\u52d9\u4e0a\u53d6 \u5f97\u5de8\u5927\u7684\u9032\u6b65", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "\u70ba\u76ee\u6a19\u51fd\u6578\u5c0d\u6a21\u578b\u9032\u884c\u8abf\u6574\u7684\u60f3\u6cd5\uff0c \u4e26\u5be6\u4f5c\u65bc\u6df1\u5c64\u985e\u795e\u7d93\u7db2\u8def\u8072\u5b78\u6a21\u578b\u7684\u67b6\u69cb\u4e0a\u63a2\u8a0e\u5c0d\u65bc\u932f\u8aa4\u767c\u97f3\u6aa2\u6e2c\u4efb\u52d9\u7684\u5f71\u97ff\u3002 \u672c\u7bc7\u8ad6\u6587\u5728\u7b2c\u4e8c\u7bc0\u5c07\u4ecb\u7d39\u932f\u8aa4\u767c\u97f3\u6aa2\u6e2c\u76f8\u95dc\u7814\u7a76\u7684\u767c\u5c55\u8fd1\u6cc1\uff1b\u7b2c\u4e09\u7bc0\u7c21\u55ae\u7684\u56de\u9867\u932f \u8aa4\u767c\u97f3\u6aa2\u6e2c\u4efb\u52d9\u4e2d\u8f03\u5e38\u88ab\u4f7f\u7528\u7684\u65b9\u6cd5\uff1b\u7b2c\u56db\u7bc0\u5247\u662f\u8a0e\u8ad6\u57fa\u65bc\u7b2c\u4e09\u7bc0\u7684\u767c\u97f3\u6aa2\u6e2c\u65b9\u6cd5\u5982\u4f55 \u5be6\u73fe\u6700\u5927\u5316\u932f\u8aa4\u767c\u97f3\u6aa2\u6e2cF \u5ea6\u91cf\u4e4b\u8a13\u7df4\uff1b\u7b2c\u4e94\u7bc0\u5247\u662f\u5f9e\u5be6\u9a57\u4e2d\u63a2\u8a0e\u6700\u5927\u5316\u932f\u8aa4\u767c\u97f3\u6aa2\u6e2c F \u5ea6\u91cf\u4e4b\u8a13\u7df4\u5c0d\u65bc\u767c\u97f3\u6aa2\u6e2c\u4efb\u52d9\u7684\u5f71\u97ff\uff1b\u6700\u5f8c\uff0c\u5728\u7b2c\u516d\u7bc0\uff0c\u6211\u5011\u63d0\u51fa\u7d50\u8ad6\u8207\u4e00\u4e9b\u672a\u4f86\u53ef \u80fd\u7684\u7814\u7a76\u65b9\u5411\u3002 2. \u6587\u737b\u63a2\u8a0e \u932f \u8aa4 \u767c \u97f3 \u6aa2 \u6e2c \u5927 \u81f4 \u53ef \u5206 \u70ba \u57fa \u65bc \u9580 \u6abb \u503c (Thresholding-Based) \u8207 \u57fa \u65bc \u5206 \u985e \u5668 (Classification-Based)\u7b49\u5169\u7a2e\u505a\u6cd5\u3002\u5169\u8005\u5dee\u5225\u5728\u65bc\u662f\u5426\u4f7f\u7528\u660e\u78ba\u7684\u9580\u6abb\u503c\u4f86\u5224\u65b7\u767c\u97f3\u70ba\u6b63 \u78ba\u6216\u932f\u8aa4\uff1b\u57fa\u65bc\u5206\u985e\u5668\u5247\u662f\u6574\u5408\u591a\u7a2e\u7279\u5fb5\u4e26\u8a13\u7df4\u4e8c\u5143\u5206\u985e\u5668\u4f86\u6c7a\u5b9a\u767c\u97f3\u662f\u5426\u5408\u683c\u3002\u57fa\u65bc \u9580 \u6abb \u503c \u7b49 \u65b9 \u6cd5 \u65e9 \u671f \u7531 (Hsu et al., 2016) \u63d0 \u51fa \u4e09 \u7a2e \u767c \u97f3 \u6aa2 \u6e2c \u7279 \u5fb5 \uff1a \u5c0d \u6578 \u76f8 \u4f3c \u5ea6 \u503c (Log-Likelihood) \u3001 \u5c0d \u6578 \u4e8b \u5f8c \u6a5f \u7387 (Log Posterior Probability) \u3001 \u6bb5 \u843d \u5340 \u9593 \u9577 \u5ea6 (Segment \u8a31\u66dc\u9e92 \u7b49 Duration)\u5c0d\u65bc\u767c\u97f3\u6aa2\u6e2c\u6548\u679c\u7684\u5f71\u97ff\u3002\u5b78\u8005 Kim \u5728\u5be6\u9a57\u4e2d\u6307\u51fa\u5c0d\u6578\u4e8b\u5f8c\u6a5f\u7387\u70ba\u8868\u73fe\u8f03\u597d\u7684 \u767c\u97f3\u6aa2\u6e2c\u5206\u6578", | |
| "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": "\u7684\u89c0\u5bdf\u4e26\u5c07\u4e8b\u5f8c\u6a5f\u7387\u6539\u826f\u4e26\u7a31\u4f5c GOP (Witt & Young, 2000)\uff0c\u4e5f \u662f\u6700\u5e38\u88ab\u4f7f\u7528\u7684\u767c\u97f3\u6aa2\u6e2c\u65b9\u6cd5\u3002GOP \u7684\u8a08\u7b97\u65b9\u5f0f\u5982\u4e0b\uff1a GOP , \u2261 , log , | , (1) , log , | , , \u2211 , | \u2208 ,", | |
| "eq_num": "(2)" | |
| } | |
| ], | |
| "section": "", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": ", log , | , \u2208 , , |", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "(3) (Hu et al., 2015; Qian et al., 2012; Hu et al., 2014) ", | |
| "cite_spans": [ | |
| { | |
| "start": 4, | |
| "end": 21, | |
| "text": "(Hu et al., 2015;", | |
| "ref_id": "BIBREF8" | |
| }, | |
| { | |
| "start": 22, | |
| "end": 40, | |
| "text": "Qian et al., 2012;", | |
| "ref_id": "BIBREF17" | |
| }, | |
| { | |
| "start": 41, | |
| "end": 57, | |
| "text": "Hu et al., 2014)", | |
| "ref_id": "BIBREF7" | |
| } | |
| ], | |
| "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": "\u5176\u4e2d GOP \u662f\u97f3\u7d20\u6bb5\u843d , \u5c0d\u61c9\u76ee\u6a19\u97f3\u7d20 , \u7684\u4e8b\u5f8c\u6a5f\u7387\uff0c\u5176\u4e2d u \u8207 n \u8868\u793a\u7b2c u \u500b\u8a9e\u53e5\u7684 \u7b2c n \u500b\u97f3\u7d20\uff0c\u6839\u64da\u8c9d\u6c0f\u5b9a\u7406\u5c07\u5f0f(1)\u8f49\u63db\u6210\u5f0f(2)\uff1b , \u662f\u8a72\u6bb5\u843d\u5c0d\u61c9\u7684\u97f3\u7d20\u96c6\u5408\uff0c\u53ef\u4ee5\u662f \u5168\u90e8\u97f3\u7d20\u6216\u90e8\u5206\u8f03\u6df7\u6dc6\u7684\u97f3\u7d20\uff0c , \u5247\u662f\u97f3\u7d20\u6bb5\u843d\u7684\u7d93\u6b77\u6642\u9593(Duration)\u3002\u6211\u5011\u5047\u8a2d\u6bcf\u500b \u97f3\u7d20\u7684\u4e8b\u524d\u6a5f\u7387\u76f8\u540c\uff0c\u4e14\u53ea\u4f7f\u7528\u6700\u5927\u76f8\u4f3c\u5ea6\u503c\u7684\u97f3\u7d20\uff0c\u5373\u6700\u6df7\u6dc6\u97f3\u7d20\u505a\u70ba\u5206\u6bcd\u9805\uff0c\u5982\u5f0f (3)\u3002\u5176\u4e2d , | , \u662f\u5df2\u77e5\u97f3\u7d20 , \u8981\u53d6\u5f97\u97f3\u7d20\u6bb5\u843d , \u7684\u76f8\u4f3c\u5ea6\u503c\uff0c\u8a08\u7b97 , | , \u53ef \u4ee5 \u900f \u904e \u5df2 \u77e5 \u7684 \u6587 \u672c \u5167 \u5bb9 \u5c0d \u8a9e \u53e5 \u9032 \u884c \u5f37 \u5236 \u5c0d \u4f4d \u53d6 \u5f97 \u5c0d \u61c9 \u97f3 \u7d20 , \u7684 \u72c0 \u614b \u5e8f \u5217 * , , \u2026 , \uff0c\u540c\u6642\u4e5f\u53ef\u4ee5\u5f97\u5230\u97f3\u7d20\u6bb5\u843d\u5340\u9593\u5c0d\u61c9\u7684\u8d77\u59cb\u6642\u9593 \u8207\u7d50\u675f\u6642\u9593 \u3002\u5f0f(3) \u6240\u8a08\u7b97\u7684 GOP \u5206\u6578\u4f5c\u70ba\u6c7a\u7b56\u767c\u97f3\u932f\u8aa4\u8207\u5426\u7684\u8a55\u4f30\u4f9d\u64da\uff0c\u4e26\u7d93\u904e\u5f0f(3)\u6c7a\u5b9a\u767c\u97f3\u7a0b\u5ea6\u7684\u5206 \u6578\u3002\u6211\u5011\u5b9a\u7fa9\u51fd\u6578 D(\u2022)\u8868\u793a\u767c\u97f3\u7684\u6c7a\u7b56\u51fd\u6578\uff1a D , \u2022 ,", | |
| "eq_num": "(4)" | |
| } | |
| ], | |
| "section": "", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "EQUATION", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [ | |
| { | |
| "start": 0, | |
| "end": 8, | |
| "text": "EQUATION", | |
| "ref_id": "EQREF", | |
| "raw_str": "\u800c D(\u2022)\u63a5\u8fd1 1 \u8868\u793a\u767c\u97f3\u53ef\u80fd\u932f\u8aa4\uff0c\u63a5\u8fd1 0 \u5247\u8868\u793a\u767c\u97f3\u6b63\u78ba\uff0c \u8868\u793a\u6c7a\u7b56\u7528\u7684\u9580\u6abb\u503c\uff0c\u800c\u53c3 \u6578 \u7528\u4f86\u5c07 GOP \u5206\u6578\u653e\u5927\u6216\u7e2e\u5c0f\u3002\u4e0a\u8ff0\u5169\u500b\u53c3\u6578\u7686\u53ef\u4ee5\u8a2d\u8a08\u70ba\u97f3\u7d20\u76f8\u4f9d\uff0c\u82e5\u70ba\u97f3\u7d20\u76f8\u4f9d \u5247\u7528 \u8207 \u8868\u793a\u3002\u63a5\u8457\u6211\u5011\u5229\u7528\u6307\u793a\u51fd\u6578\u5224\u5b9a\u767c\u97f3\u662f\u5426\u932f\u8aa4\uff1a D , 1 if D , 0 otherwise (5) \u70ba\u5168\u57df\u7684\u56fa\u5b9a\u9580\u6abb\u503c\uff0c\u5927\u90e8\u5206\u90fd\u662f\u900f\u904e\u767c\u5c55\u96c6\u8abf\u6574\u81f3\u4e00\u500b\u8f03\u5408\u9069\u7684\u503c\u3002\u7136\u800c GOP \u662f\u932f \u8aa4\u767c\u97f3\u6aa2\u6e2c\u7684\u65b9\u6cd5\u4e2d\u8f03\u666e\u904d\u7684\u4f5c\u6cd5\uff0c\u4e14\u4e0d\u9700\u4f9d\u8cf4\u4eba\u5de5\u6a19\u8a18\u597d\u7684\u932f\u8aa4\u767c\u97f3\uff0c\u5c6c\u65bc\u975e\u76e3\u7763\u5f0f \u5b78\u7fd2(Unsupervised Learning)\u7684\u65b9\u6cd5\u3002 \u6b64\u5916\uff0c\u5df2\u6709\u5b78\u8005\u63d0\u51fa\u5229\u7528\u6df1\u5c64\u985e\u795e\u7d93\u7db2\u8def\u8072\u5b78\u6a21\u578b\u7684\u8f38\u51fa\u70ba\u4e8b\u5f8c\u6a5f\u7387 | \u7684\u65b9\u6cd5 \u4f5c\u70ba\u767c\u97f3\u6aa2\u6e2c\u7684\u5206\u6578\uff0c\u7a31\u4f5c\u5c0d\u6578\u97f3\u7d20\u4e8b\u5f8c\u6a5f\u7387(Log Phone Posterior, LPP) (Hu et al., 2015)\u3002 \u5176\u8a08\u7b97\u65b9\u5f0f\u70ba\u97f3\u7d20\u6bb5\u843d , \u5c0d\u61c9\u7684\u72c0\u614b\u4e8b\u5f8c\u6a5f\u7387\u4e4b\u5e7e\u4f55\u5e73\u5747\u3002\u8207 GOP \u7684\u7b97\u6cd5\u985e\u4f3c\uff0c\u900f\u904e \u5df2 \u77e5 \u7684 \u6587 \u672c \u5167 \u5bb9 \u5c0d \u8a9e \u53e5 \u9032 \u884c \u5f37 \u5236 \u5c0d \u4f4d \u53d6 \u5f97 \u5c0d \u61c9 \u76ee \u6a19 \u97f3 \u7d20 , \u7684 \u72c0 \u614b \u5e8f \u5217 , \u8a31\u66dc\u9e92 \u7b49 , , \u2026 , \uff0c\u800c\u8a08\u7b97 LPP \u7684\u516c\u5f0f\u53ef\u4ee5\u5beb\u6210\uff1a LPP , log , | , ; ,", | |
| "eq_num": "(6)" | |
| } | |
| ], | |
| "section": "", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "EQUATION", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [ | |
| { | |
| "start": 0, | |
| "end": 8, | |
| "text": "EQUATION", | |
| "ref_id": "EQREF", | |
| "raw_str": "\u2211 log ,", | |
| "eq_num": "(7)" | |
| } | |
| ], | |
| "section": "", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "EQUATION", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [ | |
| { | |
| "start": 0, | |
| "end": 8, | |
| "text": "EQUATION", | |
| "ref_id": "EQREF", | |
| "raw_str": "\u900f\u904e\u5f0f(7)\u7b97\u51fa\u76ee\u6a19\u97f3\u7d20 , \u7684 LPP\uff0c , \u70ba\u97f3\u7d20 , \u5728\u97f3\u7d20\u6bb5\u843d , \u7684\u6700\u4f73\u8def\u5f91\u6240\u5c0d\u61c9 \u7684\u72c0\u614b\u5e8f\u5217\u3002\u5f9e\u6211\u5011\u5be6\u9a57\u4e2d\u53ef\u4ee5\u767c\u73fe\u4f7f\u7528 LPP \u7522\u751f\u767c\u97f3\u5206\u6578\u5728\u767c\u97f3\u6aa2\u6e2c\u4efb\u52d9\u7684\u6548\u679c\u8207 GOP \u76f8\u8fd1\uff0c\u4f46 LPP \u7684\u8a08\u7b97\u8907\u96dc\u5ea6\u9060\u4f4e\u65bc GOP\u3002\u5982\u5f0f(3)\u6240\u898b\uff0cGOP \u5728\u5206\u6bcd\u9805\u9700\u8981\u5c07\u6240\u6709 \u97f3\u7d20\u7684\u76f8\u4f3c\u5ea6\u503c\u7b97\u51fa\uff1b\u800c LPP \u53ea\u9700\u8981\u8a08\u7b97\u76ee\u6a19\u97f3\u7d20 , \u7684\u72c0\u614b\u4e8b\u5f8c\u6a5f\u7387\u4e4b\u5e7e\u4f55\u5e73\u5747\uff0c\u7b26 \u5408\u6df1\u5c64\u985e\u795e\u7d93\u7db2\u8def\u67b6\u69cb\u7684\u8f38\u51fa\u72c0\u614b\u4e8b\u5f8c\u6a5f\u7387\u3002\u7576\u6211\u5011\u53d6\u5f97\u4ee5 LPP \u8868\u793a\u7684\u767c\u97f3\u5206\u6578\u5f8c\uff0c\u5beb \u6210\u6c7a\u7b56\u51fd\u6578\u7684\u5f62\u5f0f\u5247\u70ba\uff1a D , \u2022 ,", | |
| "eq_num": "(8)" | |
| } | |
| ], | |
| "section": "", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "EQUATION", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [ | |
| { | |
| "start": 0, | |
| "end": 8, | |
| "text": "EQUATION", | |
| "ref_id": "EQREF", | |
| "raw_str": "\u63a5\u8457\u5c07\u6211\u5011\u5728\u7b2c 3 \u7bc0\u5b9a\u7fa9\u7684\u932f\u8aa4\u767c\u97f3\u6c7a\u7b56\u51fd\u6578 D(\u2022)\u900f\u904e\u6307\u793a\u51fd\u6578 \u2022 \u8f49\u6210\u975e 1 \u5373 0 \u7684 \u6578\u503c\uff0c\u8a13\u7df4\u8cc7\u6599\u7684\u6240\u6709\u97f3\u7d20\u6bb5\u843d\u7d93\u904e\u6c7a\u7b56\u51fd\u6578 D(\u2022)\u8207\u6307\u793a\u51fd\u6578 \u2022 \u7684\u7e3d\u548c\u70ba \uff1b\u6bcf\u500b\u97f3\u7d20 \u6bb5\u843d\u7684\u6c7a\u7b56\u8207\u5c08\u5bb6\u8a55\u65b7\u4e4b\u7d50\u679c H(\u2022)\u76f8\u4e58\u7684\u7e3d\u548c\u5247\u70ba \u2229 \uff0c\u5982\u5f0f(13)\uff1a \u2211 \u2211 , \u2022 , \u2211 \u2211 ,", | |
| "eq_num": "(13)" | |
| } | |
| ], | |
| "section": "", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "EQUATION", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [ | |
| { | |
| "start": 0, | |
| "end": 8, | |
| "text": "EQUATION", | |
| "ref_id": "EQREF", | |
| "raw_str": "\u7136\u800c\u4e0a\u8ff0\u5b9a\u7fa9\u7684F \u5ea6\u91cf\u4e26\u4e0d\u662f\u53ef\u5fae\u5206\u7684\u51fd\u6578\uff0c\u56e0\u70ba\u5728\u8a08\u7b97 \u2229 \u8207 \u6642\u4f7f\u7528\u5230\u7684\u6307\u793a\u51fd\u6578 \u2022 \u5728\u57fa\u65bc\u68af\u5ea6\u6cd5(Gradient Based Method)\u7684\u53c3\u6578\u66f4\u65b0\u65b9\u5f0f\u4e2d\u8f03\u96e3\u8655\u7406\u3002\u56e0\u6b64\u6211\u5011\u5b9a\u7fa9\u4e00 \u500b\u5e73\u6ed1(Smooth)\u7684F \u5ea6\u91cf\uff0c\u5982\u5f0f(14)\uff1a \u039e \u2022 \u2229 (14) \u2211 \u2211 , \u2022 , \u2211 \u2211 ,", | |
| "eq_num": "(15)" | |
| } | |
| ], | |
| "section": "", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "EQUATION", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [ | |
| { | |
| "start": 0, | |
| "end": 8, | |
| "text": "EQUATION", | |
| "ref_id": "EQREF", | |
| "raw_str": "\u039e \u2211 \u2211 , \u2022 , \u2211 \u2211 ,", | |
| "eq_num": "(22)" | |
| } | |
| ], | |
| "section": "", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "Bergstra, J., Breuleux, O., Bastien, F., Lamblin, P., Pascanu, R., Desjardins, G., Turian, J., Warde-Farley, D. & Bengio, Y. (2010) . Theano: A CPU and GPU math compiler in Python. In Proceedings of the Python for Scientific Computing Conference (SciPy), 1-7.", | |
| "cite_spans": [ | |
| { | |
| "start": 95, | |
| "end": 131, | |
| "text": "Warde-Farley, D. & Bengio, Y. (2010)", | |
| "ref_id": null | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "\u53c3\u8003\u6587\u737b References", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "Chen, L. Y. & Jang, J. S. R. (2015). Automatic pronunciation scoring with score combination by learning to rank and class-normalized DP-based quantization. IEEE Transactions on Audio, Speech, and Language Processing, 23(11) , 1737-1749.", | |
| "cite_spans": [ | |
| { | |
| "start": 177, | |
| "end": 183, | |
| "text": "Audio,", | |
| "ref_id": null | |
| }, | |
| { | |
| "start": 184, | |
| "end": 191, | |
| "text": "Speech,", | |
| "ref_id": null | |
| }, | |
| { | |
| "start": 192, | |
| "end": 216, | |
| "text": "and Language Processing,", | |
| "ref_id": null | |
| }, | |
| { | |
| "start": 217, | |
| "end": 223, | |
| "text": "23(11)", | |
| "ref_id": null | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "\u53c3\u8003\u6587\u737b References", | |
| "sec_num": null | |
| } | |
| ], | |
| "back_matter": [ | |
| { | |
| "text": "Wei, S., Hu, G., Hu, Y. & Wang, R. H. (2009) . A new method for mispronunciation detection using support vector machine based on pronunciation space models. ", | |
| "cite_spans": [ | |
| { | |
| "start": 9, | |
| "end": 44, | |
| "text": "Hu, G., Hu, Y. & Wang, R. H. (2009)", | |
| "ref_id": null | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "\u8a31\u66dc\u9e92 \u7b49", | |
| "sec_num": null | |
| } | |
| ], | |
| "bib_entries": { | |
| "BIBREF0": { | |
| "ref_id": "b0", | |
| "title": "An exact algorithm for F-measure maximization", | |
| "authors": [ | |
| { | |
| "first": "K", | |
| "middle": [ | |
| "J" | |
| ], | |
| "last": "Dembczynski", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "W", | |
| "middle": [], | |
| "last": "Waegeman", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "W", | |
| "middle": [], | |
| "last": "Cheng", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "E", | |
| "middle": [], | |
| "last": "H\u00fcllermeier", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2011, | |
| "venue": "Advances in Neural Information Processing Systems", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "1404--1412", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Dembczynski, K. J., Waegeman, W., Cheng, W. & H\u00fcllermeier, E. (2011). An exact algorithm for F-measure maximization. In Advances in Neural Information Processing Systems, 1404-1412.", | |
| "links": null | |
| }, | |
| "BIBREF1": { | |
| "ref_id": "b1", | |
| "title": "Multi-label Text Categorization with Model Combination based on F1-score Maximization", | |
| "authors": [ | |
| { | |
| "first": "A", | |
| "middle": [], | |
| "last": "Fujino", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "H", | |
| "middle": [], | |
| "last": "Isozaki", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "J", | |
| "middle": [], | |
| "last": "Suzuki", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2008, | |
| "venue": "Proceedings of the International Joint Conference on Natural Language Processing (IJCNLP)", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "823--828", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Fujino, A., Isozaki, H. & Suzuki, J. (2008). Multi-label Text Categorization with Model Combination based on F1-score Maximization. In Proceedings of the International Joint Conference on Natural Language Processing (IJCNLP), 823-828.", | |
| "links": null | |
| }, | |
| "BIBREF2": { | |
| "ref_id": "b2", | |
| "title": "Improving mispronunciation detection and diagnosis of learners' speech with context-sensitive phonological rules based on language transfer", | |
| "authors": [ | |
| { | |
| "first": "A", | |
| "middle": [ | |
| "M" | |
| ], | |
| "last": "Harrison", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "W", | |
| "middle": [ | |
| "Y" | |
| ], | |
| "last": "Lau", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "H", | |
| "middle": [ | |
| "M" | |
| ], | |
| "last": "Meng", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "L", | |
| "middle": [], | |
| "last": "Wang", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2008, | |
| "venue": "Proceedings of the International Conference on Speech Communication and Technology (INTERSPEECH)", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "2787--2790", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Harrison, A. M., Lau, W. Y., Meng, H. M. & Wang, L. (2008). Improving mispronunciation detection and diagnosis of learners' speech with context-sensitive phonological rules based on language transfer. In Proceedings of the International Conference on Speech Communication and Technology (INTERSPEECH), 2787-2790.", | |
| "links": null | |
| }, | |
| "BIBREF3": { | |
| "ref_id": "b3", | |
| "title": "Implementation of an extended recognition network for mispronunciation detection and diagnosis in computer-assisted pronunciation training", | |
| "authors": [ | |
| { | |
| "first": "A", | |
| "middle": [ | |
| "M" | |
| ], | |
| "last": "Harrison", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "W", | |
| "middle": [ | |
| "K" | |
| ], | |
| "last": "Lo", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "X", | |
| "middle": [], | |
| "last": "Qian", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "H", | |
| "middle": [], | |
| "last": "Meng", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2009, | |
| "venue": "Proceedings of the International Symposium on Languages, Applications and Technologies (SLaTE)", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "45--48", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Harrison, A. M., Lo, W. K., Qian, X. & Meng, H. (2009). Implementation of an extended recognition network for mispronunciation detection and diagnosis in computer-assisted pronunciation training. In Proceedings of the International Symposium on Languages, Applications and Technologies (SLaTE), 45-48.", | |
| "links": null | |
| }, | |
| "BIBREF4": { | |
| "ref_id": "b4", | |
| "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": [ | |
| "R" | |
| ], | |
| "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": [], | |
| "last": "Sainath", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "B", | |
| "middle": [], | |
| "last": "Kingsbury", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2012, | |
| "venue": "IEEE Signal Processing Magazine", | |
| "volume": "29", | |
| "issue": "6", | |
| "pages": "82--97", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Hinton, G., Deng, L., Yu, D., Dahl, G. E., Mohamed, A. R., Jaitly, N., Senior, A., Vanhoucke, V., Nguyen, P., Sainath, T. & Kingsbury, B. (2012). Deep neural networks for acoustic modeling in speech recognition: The shared views of four research groups. IEEE Signal Processing Magazine, 29(6), 82-97.", | |
| "links": null | |
| }, | |
| "BIBREF5": { | |
| "ref_id": "b5", | |
| "title": "Development of Mandarin Annotated Spoken Corpus (MAS Corpus) and the Learner Corpus Analysis", | |
| "authors": [ | |
| { | |
| "first": "Y", | |
| "middle": [], | |
| "last": "Hsiung", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Y", | |
| "middle": [], | |
| "last": "Sung", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2014, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Hsiung, Y. & Sung, Y. (2014). Development of Mandarin Annotated Spoken Corpus (MAS Corpus) and the Learner Corpus Analysis. 1st Workshop on the Analysis of Linguistic Features (WoALF).", | |
| "links": null | |
| }, | |
| "BIBREF6": { | |
| "ref_id": "b6", | |
| "title": "Mispronunciation detection leveraging maximum performance criterion training of acoustic models and decision functions", | |
| "authors": [ | |
| { | |
| "first": "Y", | |
| "middle": [ | |
| "C" | |
| ], | |
| "last": "Hsu", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "M", | |
| "middle": [ | |
| "H" | |
| ], | |
| "last": "Yang", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "H", | |
| "middle": [ | |
| "T" | |
| ], | |
| "last": "Hung", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "B", | |
| "middle": [], | |
| "last": "Chen", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2016, | |
| "venue": "Proceedings of the International Conference on Speech Communication and Technology (INTERSPEECH)", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "2646--2650", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Hsu, Y. C., Yang, M. H., Hung, H. T. & Chen, B. (2016). Mispronunciation detection leveraging maximum performance criterion training of acoustic models and decision functions. In Proceedings of the International Conference on Speech Communication and Technology (INTERSPEECH), 2646-2650.", | |
| "links": null | |
| }, | |
| "BIBREF7": { | |
| "ref_id": "b7", | |
| "title": "A DNN-based acoustic modeling of tonal language and its application to Mandarin pronunciation training", | |
| "authors": [ | |
| { | |
| "first": "W", | |
| "middle": [], | |
| "last": "Hu", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Y", | |
| "middle": [], | |
| "last": "Qian", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "F", | |
| "middle": [ | |
| "K" | |
| ], | |
| "last": "Soong", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2014, | |
| "venue": "Proceedings of the International Conference on Acoustics Speech and Signal Processing", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "3206--3210", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Hu, W., Qian, Y. & Soong, F. K. (2014). A DNN-based acoustic modeling of tonal language and its application to Mandarin pronunciation training. In Proceedings of the International Conference on Acoustics Speech and Signal Processing (ICASSP), 3206-3210.", | |
| "links": null | |
| }, | |
| "BIBREF8": { | |
| "ref_id": "b8", | |
| "title": "Improved mispronunciation detection with deep neural network trained acoustic models and transfer learning based logistic regression classifiers", | |
| "authors": [ | |
| { | |
| "first": "W", | |
| "middle": [], | |
| "last": "Hu", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Y", | |
| "middle": [], | |
| "last": "Qian", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "F", | |
| "middle": [ | |
| "K" | |
| ], | |
| "last": "Soong", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Y", | |
| "middle": [], | |
| "last": "Wang", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2015, | |
| "venue": "Speech Communication", | |
| "volume": "67", | |
| "issue": "", | |
| "pages": "154--166", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Hu, W., Qian, Y., Soong, F. K. & Wang, Y. (2015). Improved mispronunciation detection with deep neural network trained acoustic models and transfer learning based logistic regression classifiers. Speech Communication, 67, 154-166.", | |
| "links": null | |
| }, | |
| "BIBREF9": { | |
| "ref_id": "b9", | |
| "title": "Maximum F1-score discriminative training for automatic mispronunciation detection in computer-assisted language learning", | |
| "authors": [ | |
| { | |
| "first": "H", | |
| "middle": [], | |
| "last": "Huang", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "J", | |
| "middle": [], | |
| "last": "Wang", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "H", | |
| "middle": [], | |
| "last": "Abudureyimu", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2012, | |
| "venue": "Proceedings of the International Conference on Speech Communication and Technology (INTERSPEECH)", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "815--818", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Huang, H., Wang, J. & Abudureyimu, H. (2012). Maximum F1-score discriminative training for automatic mispronunciation detection in computer-assisted language learning. In Proceedings of the International Conference on Speech Communication and Technology (INTERSPEECH), 815-818.", | |
| "links": null | |
| }, | |
| "BIBREF10": { | |
| "ref_id": "b10", | |
| "title": "Maximum F1-score discriminative training criterion for automatic mispronunciation detection", | |
| "authors": [ | |
| { | |
| "first": "H", | |
| "middle": [], | |
| "last": "Huang", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "H", | |
| "middle": [], | |
| "last": "Xu", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "X", | |
| "middle": [], | |
| "last": "Wang", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "W", | |
| "middle": [], | |
| "last": "Silamu", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2015, | |
| "venue": "IEEE Transactions on Audio, Speech, and Language Processing", | |
| "volume": "23", | |
| "issue": "4", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Huang, H., Xu, H., Wang, X. & Silamu, W. (2015). Maximum F1-score discriminative training criterion for automatic mispronunciation detection. IEEE Transactions on Audio, Speech, and Language Processing, 23(4), 787-797. \u8a55\u4f30\u5c3a\u5ea6\u76f8\u95dc\u6700\u4f73\u5316\u65b9\u6cd5\u65bc\u83ef\u8a9e\u932f\u8aa4\u767c\u97f3\u6aa2\u6e2c\u4e4b\u7814\u7a76 69", | |
| "links": null | |
| }, | |
| "BIBREF11": { | |
| "ref_id": "b11", | |
| "title": "Automatic pronunciation scoring of specific phone segments for language instruction", | |
| "authors": [ | |
| { | |
| "first": "Y", | |
| "middle": [], | |
| "last": "Kim", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "H", | |
| "middle": [], | |
| "last": "Franco", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "L", | |
| "middle": [], | |
| "last": "Neumeyer", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1997, | |
| "venue": "Proceedings of the European Conference on Speech Communication and Technology", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "649--652", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Kim, Y., Franco, H. & Neumeyer, L. (1997). Automatic pronunciation scoring of specific phone segments for language instruction. In Proceedings of the European Conference on Speech Communication and Technology (EUROSPEECH), 649-652.", | |
| "links": null | |
| }, | |
| "BIBREF12": { | |
| "ref_id": "b12", | |
| "title": "Pronunciation assessment of Japanese learners of French with GOP scores and phonetic information", | |
| "authors": [ | |
| { | |
| "first": "V", | |
| "middle": [], | |
| "last": "Laborde", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "T", | |
| "middle": [], | |
| "last": "Pellegrini", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "L", | |
| "middle": [], | |
| "last": "Fontan", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "J", | |
| "middle": [], | |
| "last": "Mauclair", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "H", | |
| "middle": [], | |
| "last": "Sahraoui", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "J", | |
| "middle": [], | |
| "last": "Farinas", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2016, | |
| "venue": "Proceedings of the International Conference on Speech Communication and Technology (INTERSPEECH)", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "2686--2690", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Laborde, V., Pellegrini, T., Fontan, L., Mauclair, J., Sahraoui, H. & Farinas, J. (2016). Pronunciation assessment of Japanese learners of French with GOP scores and phonetic information. In Proceedings of the International Conference on Speech Communication and Technology (INTERSPEECH), 2686-2690.", | |
| "links": null | |
| }, | |
| "BIBREF13": { | |
| "ref_id": "b13", | |
| "title": "A comparison-based approach to mispronunciation detection", | |
| "authors": [ | |
| { | |
| "first": "A", | |
| "middle": [], | |
| "last": "Lee", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "J", | |
| "middle": [], | |
| "last": "Glass", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2012, | |
| "venue": "Proceedings of the International Conference on Spoken Language Technology Workshop (SLT)", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "382--387", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Lee, A. & Glass, J. (2012). A comparison-based approach to mispronunciation detection. In Proceedings of the International Conference on Spoken Language Technology Workshop (SLT), 382-387.", | |
| "links": null | |
| }, | |
| "BIBREF14": { | |
| "ref_id": "b14", | |
| "title": "Automatic derivation of phonological rules for mispronunciation detection in a computer-assisted pronunciation training system", | |
| "authors": [ | |
| { | |
| "first": "W", | |
| "middle": [ | |
| "K" | |
| ], | |
| "last": "Lo", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "S", | |
| "middle": [], | |
| "last": "Zhang", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "H", | |
| "middle": [ | |
| "M" | |
| ], | |
| "last": "Meng", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2010, | |
| "venue": "Proceedings of the International Conference on Speech Communication and Technology (INTERSPEECH)", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "765--768", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Lo, W. K., Zhang, S. & Meng, H. M. (2010). Automatic derivation of phonological rules for mispronunciation detection in a computer-assisted pronunciation training system. In Proceedings of the International Conference on Speech Communication and Technology (INTERSPEECH), 765-768.", | |
| "links": null | |
| }, | |
| "BIBREF15": { | |
| "ref_id": "b15", | |
| "title": "Scikit-learn: Machine learning in Python", | |
| "authors": [ | |
| { | |
| "first": "F", | |
| "middle": [], | |
| "last": "Pedregosa", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "G", | |
| "middle": [], | |
| "last": "Varoquaux", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "A", | |
| "middle": [], | |
| "last": "Gramfort", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "V", | |
| "middle": [], | |
| "last": "Michel", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "B", | |
| "middle": [], | |
| "last": "Thirion", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "O", | |
| "middle": [], | |
| "last": "Grisel", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": ".", | |
| "middle": [ | |
| "." | |
| ], | |
| "last": "Vanderplas", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "J", | |
| "middle": [], | |
| "last": "", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2011, | |
| "venue": "Journal of Machine Learning Research", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "2825--2830", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Pedregosa, F., Varoquaux, G., Gramfort, A., Michel, V., Thirion, B., Grisel, O., ... & Vanderplas, J. (2011). Scikit-learn: Machine learning in Python. Journal of Machine Learning Research, 2825-2830.", | |
| "links": null | |
| }, | |
| "BIBREF16": { | |
| "ref_id": "b16", | |
| "title": "The Kaldi speech recognition toolkit", | |
| "authors": [ | |
| { | |
| "first": "D", | |
| "middle": [], | |
| "last": "Povey", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "A", | |
| "middle": [], | |
| "last": "Ghoshal", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "G", | |
| "middle": [], | |
| "last": "Boulianne", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "L", | |
| "middle": [], | |
| "last": "Burget", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "O", | |
| "middle": [], | |
| "last": "Glembek", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "N", | |
| "middle": [], | |
| "last": "Goel", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "M", | |
| "middle": [], | |
| "last": "Hannemann", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "P", | |
| "middle": [], | |
| "last": "Motl\u0131cek", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Y", | |
| "middle": [], | |
| "last": "Qian", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "P", | |
| "middle": [], | |
| "last": "Schwarz", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "J", | |
| "middle": [], | |
| "last": "Silovsky", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "G", | |
| "middle": [], | |
| "last": "Stemmer", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "K", | |
| "middle": [], | |
| "last": "Vesely", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2011, | |
| "venue": "Proceedings of the IEEE Automatic Speech Recognition and Understanding Workshop (ASRU)", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Povey, D., Ghoshal, A., Boulianne, G., Burget, L., Glembek, O., Goel, N., Hannemann, M., Motl\u0131cek, P., Qian, Y., Schwarz, P., Silovsky, J., Stemmer G. & Vesely, K. (2011). The Kaldi speech recognition toolkit. In Proceedings of the IEEE Automatic Speech Recognition and Understanding Workshop (ASRU).", | |
| "links": null | |
| }, | |
| "BIBREF17": { | |
| "ref_id": "b17", | |
| "title": "The Use of DBN-HMMs for Mispronunciation Detection and Diagnosis in L2 English to Support Computer-Aided Pronunciation Training", | |
| "authors": [ | |
| { | |
| "first": "X", | |
| "middle": [], | |
| "last": "Qian", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "H", | |
| "middle": [ | |
| "M" | |
| ], | |
| "last": "Meng", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "F", | |
| "middle": [ | |
| "K" | |
| ], | |
| "last": "Soong", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2012, | |
| "venue": "Proceedings of the International Conference on Speech Communication and Technology (INTERSPEECH)", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "775--778", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Qian, X., Meng, H. M. & Soong, F. K. (2012). The Use of DBN-HMMs for Mispronunciation Detection and Diagnosis in L2 English to Support Computer-Aided Pronunciation Training. In Proceedings of the International Conference on Speech Communication and Technology (INTERSPEECH), 775-778.", | |
| "links": null | |
| }, | |
| "BIBREF18": { | |
| "ref_id": "b18", | |
| "title": "Discriminative acoustic model for improving mispronunciation detection and diagnosis in computer-aided pronunciation training (CAPT)", | |
| "authors": [ | |
| { | |
| "first": "X", | |
| "middle": [], | |
| "last": "Qian", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "F", | |
| "middle": [ | |
| "K" | |
| ], | |
| "last": "Soong", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "H", | |
| "middle": [ | |
| "M" | |
| ], | |
| "last": "Meng", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2010, | |
| "venue": "Proceedings of the International Conference on Speech Communication and Technology (INTERSPEECH)", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "757--760", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Qian, X., Soong, F. K. & Meng, H. M. (2010). Discriminative acoustic model for improving mispronunciation detection and diagnosis in computer-aided pronunciation training (CAPT). In Proceedings of the International Conference on Speech Communication and Technology (INTERSPEECH), 757-760.", | |
| "links": null | |
| }, | |
| "BIBREF19": { | |
| "ref_id": "b19", | |
| "title": "Comparing classifiers for pronunciation error detection", | |
| "authors": [ | |
| { | |
| "first": "H", | |
| "middle": [], | |
| "last": "Strik", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "K", | |
| "middle": [ | |
| "P" | |
| ], | |
| "last": "Truong", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "F", | |
| "middle": [], | |
| "last": "De Wet", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "C", | |
| "middle": [], | |
| "last": "Cucchiarini", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2007, | |
| "venue": "Proceedings of the International Conference on Speech Communication and Technology (INTERSPEECH)", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "1837--1840", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Strik, H., Truong, K. P., De Wet, F. & Cucchiarini, C. (2007). Comparing classifiers for pronunciation error detection. In Proceedings of the International Conference on Speech Communication and Technology (INTERSPEECH), 1837-1840.", | |
| "links": null | |
| }, | |
| "BIBREF20": { | |
| "ref_id": "b20", | |
| "title": "Improved approaches of modeling and detecting error patterns with empirical analysis for computer-aided pronunciation training", | |
| "authors": [ | |
| { | |
| "first": "Y", | |
| "middle": [ | |
| "B" | |
| ], | |
| "last": "Wang", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "L", | |
| "middle": [ | |
| "S" | |
| ], | |
| "last": "Lee", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2012, | |
| "venue": "Proceedings of the International Conference on Acoustics Speech and Signal Processing", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "5049--5052", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Wang, Y. B. & Lee, L. S. (2012). Improved approaches of modeling and detecting error patterns with empirical analysis for computer-aided pronunciation training. In Proceedings of the International Conference on Acoustics Speech and Signal Processing (ICASSP), 5049-5052.", | |
| "links": null | |
| }, | |
| "BIBREF21": { | |
| "ref_id": "b21", | |
| "title": "Supervised detection and unsupervised discovery of pronunciation error patterns for computer-assisted language learning", | |
| "authors": [ | |
| { | |
| "first": "Y", | |
| "middle": [ | |
| "B" | |
| ], | |
| "last": "Wang", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "L", | |
| "middle": [ | |
| "S" | |
| ], | |
| "last": "Lee", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2015, | |
| "venue": "IEEE Transactions on Audio, Speech, and Language Processing", | |
| "volume": "23", | |
| "issue": "3", | |
| "pages": "564--579", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Wang, Y. B. & Lee, L. S. (2015). Supervised detection and unsupervised discovery of pronunciation error patterns for computer-assisted language learning. IEEE Transactions on Audio, Speech, and Language Processing, 23(3), 564-579.", | |
| "links": null | |
| } | |
| }, | |
| "ref_entries": { | |
| "TABREF5": { | |
| "html": null, | |
| "text": "\u6240\u793a\u3002\u7531\u8868 4 \u53ef\u4ee5\u5f97\u77e5\u57fa\u65bc DNN-HMM \u4f5c\u70ba \u8072\u5b78\u6a21\u578b\u7522\u751f GOP \u5206\u6578\u4e26\u61c9\u7528\u5728\u767c\u97f3\u6aa2\u6e2c\u4efb\u52d9\u6548\u679c\u66f4\u52dd GMM-HMM \u8072\u5b78\u6a21\u578b\u7684\u6548\u679c\u767c \u97f3\u6aa2\u6e2c\u7684F \u5ea6\u91cf\u7686\u6709\u7d04 3%\u7684\u7d55\u5c0d\u9032\u6b65(\u6b63\u78ba\u767c\u97f3\u6aa2\u6e2c\u7684F \u5ea6\u91cf\u7531 0.836 \u63d0\u5347\u81f3 0.863\uff1b\u932f \u8aa4\u767c\u97f3\u6aa2\u6e2c\u7684F \u5ea6\u91cf\u7531 0.546 \u63d0\u5347\u81f3 0.579)\u3002\u5df2\u6709\u8a31\u591a\u5b78\u8005\u5728\u5be6\u9a57\u4e2d\u8b49\u660e\u4e86\u6df1\u5c64\u5b78\u7fd2\u5728\u767c \u97f3\u6aa2\u6e2c\u4efb\u52d9\u7684\u7a81\u7834", | |
| "type_str": "table", | |
| "content": "<table><tr><td>62</td><td colspan=\"3\">\u8a55\u4f30\u5c3a\u5ea6\u76f8\u95dc\u6700\u4f73\u5316\u65b9\u6cd5\u65bc\u83ef\u8a9e\u932f\u8aa4\u767c\u97f3\u6aa2\u6e2c\u4e4b\u7814\u7a76</td><td>\u8a31\u66dc\u9e92 \u7b49 63 \u8a31\u66dc\u9e92 \u7b49</td></tr><tr><td colspan=\"4\">\u6599\u7686\u662f\u7531 1 \u81f3 4 \u4eba\u9032\u884c\u5be9\u8996\uff0c\u4e26\u63a1\u7528\u591a\u6578\u6c7a\u5224\u65b7\u767c\u97f3\u70ba\u6b63\u78ba\u6216\u932f\u8aa4\u3002\u6211\u5011\u5c07\u8a9e\u6599\u5eab\u5206\u6210 \u8a13\u7df4\u96c6\u3001\u767c\u5c55\u96c6\u8207\u6e2c\u8a66\u96c6\uff0c\u5982\u8868 1\u3002 \u8868 1. \u83ef\u8a9e\u5b78\u7fd2\u8005\u53e3\u8a9e\u8a9e\u6599\u5eab [Table 1. Statistics of the Mandarin Annotated Spoken Corpus.] \u904e GMM-HMM \u8072\u5b78\u6a21\u578b\u3002 Recall \u6b63\u78ba\u62d2\u7d55 \u7684\u500b\u6578 \u5be6\u969b\u70ba\u932f\u8aa4\u767c\u97f3\u7684\u500b\u6578 # # # (19) \u8868 2. \u81ea\u52d5\u8a9e\u97f3\u8fa8\u8b58\u5be6\u9a57\u7d50\u679c [Table 2. ASR experimental results.] \u97f3\u7bc0\u932f\u8aa4\u7387(%) \u97f3\u7d20\u932f\u8aa4\u7387(%) Precision \u6b63\u78ba\u62d2\u7d55 \u7684\u500b\u6578 \u7cfb\u7d71\u5224\u65b7\u70ba\u932f\u8aa4\u767c\u97f3\u7684\u500b\u6578 # # # (20)</td></tr><tr><td/><td colspan=\"2\">\u6642\u9593(\u5c0f\u6642) (syllable error rate, SER) \u8a9e\u8005(\u500b) 50.87 2\u2022Recall \u2022Precision GMM-HMM F1 Recall Precision</td><td>\u767c\u97f3\u932f\u8aa4\u4e4b (phone error rate, PER) \u97f3\u7d20\u6578\u91cf(\u500b) \u97f3\u7d20\u6578\u91cf(\u500b) 34.30</td><td>(21)</td></tr><tr><td colspan=\"4\">L1 DNN-HMM \u8a13\u7df4\u96c6 L2 \u7cbe\u6e96\u5ea6\u7684\u8cc7\u8a0a\u5728\u5176\u5b83\u5e38\u898b\u7684\u8a55\u4f30\u65b9\u5f0f(\u6e96\u78ba\u7387(Accuracy)\u6216 ROC \u66f2\u7dda\u7b49)\u4e2d\u4e0d\u6613\u89c0\u5bdf\uff0c 6.68 44 72,846 NA 41.71 28.14 14.04 74 107,202 24,150 \u767c\u5c55\u96c6 L1 1.4 10 14,186 NA L2 3.39 18 25,900 5,227 L1 3.21 25 32,568 NA \u4f46\u5c0d\u65bc\u932f\u8aa4\u767c\u97f3\u6aa2\u6e2c\u4efb\u52d9\u800c\u8a00\u7cbe\u6e96\u5ea6\u4e5f\u662f\u975e\u5e38\u91cd\u8981\u7684\u6307\u6a19\u4e4b\u4e00\uff0c\u56e0\u6b64\u540c\u6642\u8003\u616e\u53ec\u56de\u7387\u8207 5.3 \u8a55\u4f30\u65b9\u6cd5 \u7cbe\u6e96\u5ea6\u7684F \u5ea6\u91cf\u6307\u6a19\u70ba\u672c\u8ad6\u6587\u5f8c\u7e8c\u5be6\u9a57\u8a0e\u8ad6\u6700\u5e38\u4f7f\u7528\u7684\u8a55\u4f30\u6a19\u6e96\u3002 \u8868 3. ROC \u5206\u6790\u7684\u56db\u9805\u6307\u6a19\u5728\u767c\u97f3\u6aa2\u6e2c\u4efb\u52d9\u4e2d\u7684\u5b9a\u7fa9 [Table 3. The definition of the confusion matrix used in the mispronunciation detection task.] 5.4 \u932f\u8aa4\u767c\u97f3\u6aa2\u6e2c\u5be6\u9a57 \u6e2c\u8a66\u96c6 L2 7.49 44 55,190 14,247 \u63cf\u8ff0 \u5ef6\u7e8c 5.2 \u5c0f\u7bc0\u8a2d\u5b9a\u7684\u521d\u59cb\u8072\u5b78\u6a21\u578b(GMM-HMM \u8207 DNN-HMM)\uff0c\u4e26\u4f7f\u7528\u7b2c 3 \u7bc0\u63d0\u5230\u7684</td></tr><tr><td colspan=\"4\">5.2 \u8072\u5b78\u6a21\u578b\u8a13\u7df4 GOP \u5206\u6578(\u5f0f(3))\u4f5c\u70ba\u8a55\u4f30\u767c\u97f3\u54c1\u8cea\u7684\u7279\u5fb5\uff1b\u4e26\u4ee3\u5165\u6c7a\u7b56\u51fd\u6578(\u5f0f(4))\u8207\u6307\u793a\u51fd\u6578(\u5f0f(5))\u6aa2 \u932f\u8aa4\u7684\u63a5\u53d7 \u5be6\u969b\u4e0a\u5b78\u7fd2\u8005\u7684\u767c\u97f3\u932f\u8aa4\uff0c\u7cfb\u7d71\u537b\u8a8d\u5b9a\u70ba\u767c\u97f3\u6b63\u78ba\u3002 \u6e2c\u5b78\u7fd2\u8005\u7684\u767c\u97f3\u70ba\u6b63\u78ba\u6216\u932f\u8aa4\uff0c\u6c7a\u7b56\u51fd\u6578\u8207\u6307\u793a\u51fd\u6578\u7684\u53c3\u6578\u7686\u4f7f\u7528\u5168\u57df\u7684\u6578\u503c(\u672a\u8abf\u6574\u70ba (false acceptances, FA) \u97f3\u7d20\u76f8\u4f9d\u6216\u97f3\u7d20\u72c0\u614b\u76f8\u4f9d)\uff0c\u5176\u7d50\u679c\u5982\u8868 4</td></tr><tr><td/><td>\u932f\u8aa4\u7684\u62d2\u7d55 (false rejections, FR)</td><td colspan=\"2\">\u5be6\u969b\u4e0a\u5b78\u7fd2\u8005\u7684\u767c\u97f3\u6b63\u78ba\uff0c\u7cfb\u7d71\u537b\u5224\u65b7\u70ba\u767c\u97f3\u932f\u8aa4\u3002</td></tr><tr><td/><td>\u6b63\u78ba\u7684\u63a5\u53d7 (true acceptances, TA)</td><td colspan=\"2\">\u5be6\u969b\u4e0a\u5b78\u7fd2\u8005\u7684\u767c\u97f3\u6b63\u78ba\uff0c\u7cfb\u7d71\u4e5f\u5224\u65b7\u70ba\u767c\u97f3\u6b63\u78ba\u3002</td></tr><tr><td/><td>\u6b63\u78ba\u7684\u62d2\u7d55 (true rejections, TR)</td><td colspan=\"2\">pitch)\u7279 \u5be6\u969b\u4e0a\u5b78\u7fd2\u8005\u7684\u767c\u97f3\u932f\u8aa4\uff0c\u7cfb\u7d71\u4e5f\u8a8d\u5b9a\u70ba\u767c\u97f3\u932f\u8aa4\u3002</td></tr><tr><td colspan=\"4\">(7))\u5f97\u51fa\u6bcf\u7b46\u8a13\u7df4\u8cc7\u6599 \u5fb5\u6240\u7d44\u6210\uff1b\u4e26\u5c0d 16 \u7dad\u8a9e\u97f3\u7279\u5fb5\u53d6\u76f8\u5c0d\u7684\u4e00\u968e\u5dee\u91cf\u4fc2\u6578(Delta Coefficient)\u548c\u4e8c\u968e\u5dee\u91cf\u4fc2\u6578 \u5982\u540c\u672c\u8ad6\u6587\u5728\u7b2c 1 \u7bc0\u63d0\u53ca\u7684\uff0c\u4e8c\u5206\u985e\u554f\u984c\u6703\u6709\u56db\u7a2e\u7d50\u5c40(\u5982\u8868 3)\uff0c\u57fa\u65bc\u9019\u56db\u9805\u6307\u6a19\u53ef\u4ee5\u5ef6</td></tr><tr><td colspan=\"4\">\u7684\u767c\u97f3\u5206\u6578\uff0c\u63a5\u8457\u900f\u904e\u6c7a\u7b56\u51fd\u6578(\u5f0f(8))\u5c07\u767c\u97f3\u5206\u6578\u8f49\u6210\u6c7a\u7b56\u503c(\u503c\u57df 0 \u5230 1 \u4e4b\u9593)\u3002 (Acceleration Coefficient)\u5408\u4f75\u6210 48 \u7dad\u7684\u7279\u5fb5\u5411\u91cf\uff1b\u5176\u4e2d\u53d6\u4e00\u968e\u548c\u4e8c\u968e\u5dee\u91cf\u4fc2\u6578\u662f\u70ba\u4e86\u7372 \u4f38\u51fa\u975e\u5e38\u591a\u8b8a\u7684\u8a55\u4f30\u65b9\u5f0f\uff1b\u4f8b\u5982\u53ec\u56de\u7387\u8207\u7cbe\u6e96\u5ea6\u662f\u5206\u985e\u554f\u984c\u4e2d\u7d93\u5e38\u88ab\u4f7f\u7528\u7684\u8a55\u4f30\u65b9\u5f0f\uff0c</td></tr><tr><td colspan=\"4\">\u5f97\u8a9e\u97f3\u7279\u5fb5\u5728\u6642\u9593\u7684\u76f8\u95dc\u8cc7\u8a0a\u3002 \u800c\u53ec\u56de\u7387\u8207\u7cbe\u6e96\u5ea6\u7684\u8abf\u548c\u5e73\u5747\uff0c\u4e5f\u5c31\u662fF \u5ea6\u91cf\u66f4\u662f\u5ee3\u70ba\u4f7f\u7528\u3002\u7121\u8ad6\u662f\u6b63\u78ba\u6216\u932f\u8aa4\u767c\u97f3\u7684 (3) \u63a5\u7e8c\u6b65\u9a5f(2)\u7b97\u51fa\u7684\u6c7a\u7b56\u503c\u900f\u904e\u5f0f(15)\u7b97\u51fa\u8fd1\u4f3c\u7684F \u5ea6\u91cf\u4f5c\u70ba\u76ee\u6a19\u51fd\u6578\u4e26\u8fed\u4ee3\u7684\u8a13\u7df4 \u6c7a\u7b56\u51fd\u6578\u7684\u53c3\u6578 , \u4ee5\u53ca DNN-HMM \u8072\u5b78\u6a21\u578b\u7684\u53c3\u6578\uff0c\u800c\u6c7a\u7b56\u51fd\u6578\u7684\u53c3\u6578\u53ef\u70ba\u97f3\u7d20 \u76f8\u4f9d\u3002 \u6aa2\u6e2c\u7d50\u679c\u90fd\u662f\u91cd\u8981\u7684\u6307\u6a19\uff0c\u56e0\u6b64\u6211\u5011\u5148\u5b9a\u7fa9\u6b63\u78ba\u767c\u97f3\u6aa2\u6e2c\u7684\u53ec\u56de\u7387(Recall )\u3001\u7cbe\u6e96\u5ea6 \u5728 DNN-HMM \u8072\u5b78\u6a21\u578b\u7684\u90e8\u5206\uff0c\u6bcf\u500b\u96b1\u85cf\u5c64\u7686\u4f7f\u7528\u908f\u8f2f\u51fd\u6578(sigmoid function)\u4f5c\u70ba \u6fc0\u767c\u51fd\u6578\uff0c\u5230\u8f38\u51fa\u5c64\u5247\u4f7f\u7528\u8edf\u5f0f\u6700\u5927\u5316\u51fd\u6578(Softmax)\u8f49\u63db\u6210\u6a5f\u7387\u3002\u8f38\u5165\u7279\u5fb5\u662f\u6885\u723e\u983b\u8b5c (Precision )\u8207F \u5ea6\u91cf(F1 )\u7684\u8a08\u7b97\u65b9\u5f0f\uff1a</td></tr><tr><td colspan=\"4\">\u76f8\u8f03\u65bc\u539f\u672c\u7684\u6d41\u7a0b\u5728\u8a13\u7df4\u8cc7\u6599\u4e2d\u52a0\u5165\u4e86\u4e8c\u8a9e(L2)\u7684\u8cc7\u6599(\u5305\u542b\u6b63\u78ba\u8207\u932f\u8aa4\u767c\u97f3)\uff1b\u4e14\u6c7a\u7b56\u51fd \u6578\u8207\u8072\u5b78\u6a21\u578b\u7684\u53c3\u6578\u4e5f\u4ee5\u767c\u97f3\u6aa2\u6e2c\u4efb\u52d9\u7684\u76ee\u6a19\u51fd\u6578\u9032\u884c\u8abf\u9069\u3002 \u4fc2\u6578(Mel-Scale Frequency Spectral Coefficients, MFSC)\u53d6\u5f97\u7684\u5c0d\u6578\u80fd\u91cf\u7279\u5fb5\u4e26\u900f\u904e\u6ffe\u6ce2 \u5668\u7d44(Filter Banks)\u6240\u7522\u751f\u7684 40 \u7dad\u8f38\u51fa\uff1b\u9130\u8fd1\u97f3\u7a97\u6211\u5011\u63a1\u7528\u524d\u5f8c\u5404 5 \u500b\u97f3\u6846\uff0c\u5171\u542b 11 \u500b Recall \u6b63\u78ba\u63a5\u53d7 \u7684\u500b\u6578 \u5be6\u969b\u70ba\u6b63\u78ba\u767c\u97f3\u7684\u500b\u6578 # # # (16)</td></tr><tr><td colspan=\"4\">5. \u5be6\u9a57\u7d50\u679c \u97f3\u6846\uff0c\u6bcf\u500b\u97f3\u6846\u7686\u70ba 40 \u7dad\u7684\u6ffe\u6ce2\u5668\u7d44\u7522\u751f\u52a0\u4e0a 3 \u7dad\u5ea6\u97f3\u8abf(Pitch)\u7279\u5fb5\uff1b\u4e26\u5c0d 43 \u7dad\u8a9e\u97f3 \u7279\u5fb5\u53d6\u76f8\u5c0d\u7684\u4e00\u7686\u5dee\u91cf\u4fc2\u6578\u548c\u4e8c\u7686\u5dee\u91cf\u4fc2\u6578\uff0c\u5247\u8f38\u5165\u7684\u8a9e\u97f3\u7279\u5fb5\u5c31\u6703\u5f97\u5230 11 \u500b 129 \u7dad\u7684 \u7279\u5fb5\u5411\u91cf\u4e32\u6210\u4e00\u500b 1,419 \u7dad\u5ea6\u7684\u7279\u5fb5\u5411\u91cf\u3002 Precision \u6b63\u78ba\u63a5\u53d7 \u7684\u500b\u6578 \u7cfb\u7d71\u5224\u65b7\u70ba\u6b63\u78ba\u767c\u97f3\u7684\u500b\u6578 # # # (17)</td></tr><tr><td colspan=\"4\">5.1 \u83ef\u8a9e\u5b78\u7fd2\u8005\u53e3\u8a9e\u8a9e\u6599\u5eab (Phone Error Rate, PER)\u4f86\u8868\u793a\uff0c\u5982\u8868 2\uff1b\u89e3\u78bc\u904e\u7a0b\u70ba\u81ea\u7531\u97f3\u7bc0\u89e3\u78bc(Free-Syllable Decoding) \u672c\u8ad6\u6587\u4f7f\u7528\u81fa\u7063\u5e2b\u7bc4\u5927\u5b78\u9081\u5411\u9802\u5c16\u5927\u5b78\u8a08\u756b\u6240\u63d0\u4f9b\u7684\u83ef\u8a9e\u5b78\u7fd2\u8005\u53e3\u8a9e\u8a9e\u6599\u5eab(Hsiung & \u5728\u81ea\u52d5\u8a9e\u97f3\u8fa8\u8b58\u7684\u7d50\u679c\u6211\u5011\u4ee5\u97f3\u7bc0\u932f\u8aa4\u7387(Syllable Error Rate, SER)\u8207\u97f3\u7d20\u932f\u8aa4\u7387 F1 \u2022 \u2022 (18)</td></tr><tr><td colspan=\"4\">Sung, 2014)\uff0c\u8a9e\u8005\u90e8\u5206\u5305\u542b\u83ef\u8a9e\u6bcd\u8a9e\u8005(L1)\u8207\u83ef\u8a9e\u975e\u6bcd\u8a9e\u8005(L2)\uff0c\u9304\u97f3\u5167\u5bb9\u6709\u55ae\u97f3\u7bc0\u3001\u96d9 \u4e14\u7121\u4efb\u4f55\u8a9e\u8a00\u6a21\u578b\u9650\u5236\uff0c\u8fa8\u8b58\u932f\u8aa4\u7387\u662f\u4f7f\u7528\u8868 1 \u7684 L1 \u6e2c\u8a66\u96c6\u6240\u8a08\u7b97\u7684\u3002\u5f9e\u8868 2 \u7684\u8fa8\u8b58 \u800c\u932f\u8aa4\u767c\u97f3\u6aa2\u6e2c\u7684\u53ec\u56de\u7387(Recall )\u3001\u7cbe\u6e96\u5ea6(Precision )\u8207F \u5ea6\u91cf(F1 )\u7684\u8a08\u7b97\u65b9\u5f0f</td></tr><tr><td colspan=\"4\">\u97f3\u7bc0\u3001\u591a\u97f3\u7bc0\u8207\u77ed\u6587\u7b49\u60c5\u5883\uff1b\u5176\u4e2d\u83ef\u8a9e\u975e\u6bcd\u8a9e\u8005\u8a9e\u6599\u5eab\u70ba\u97f3\u7d20\u5c64\u6b21\u7684\u767c\u97f3\u6a19\u8a18\uff0c\u6bcf\u7b46\u8cc7 \u7d50\u679c\u53ef\u4ee5\u89c0\u5bdf\u5230\u7121\u8ad6\u662f\u97f3\u7bc0\u932f\u8aa4\u7387\u6216\u662f\u97f3\u7d20\u932f\u8aa4\u7387\u7686\u7531 DNN-HMM \u8072\u5b78\u6a21\u578b\u5927\u5e45\u5ea6\u52dd \u70ba\uff1a</td></tr></table>", | |
| "num": null | |
| }, | |
| "TABREF6": { | |
| "html": null, | |
| "text": "LPP \u7684\u65b9\u6cd5\u5728F \u5ea6\u91cf\u7684\u8868\u73fe\u76f8\u8fd1(\u6b63\u78ba\u767c\u97f3\u6aa2\u6e2c\u7684F \u5ea6\u91cf\u7531 0.863 \u964d\u4f4e\u81f3 0.854\uff1b\u932f\u8aa4\u767c\u97f3\u6aa2\u6e2c\u7684F \u5ea6\u91cf\u7531 0.579 \u63d0\u5347\u81f3 0.587)\uff0c\u5176\u8b8a\u5316\u7686\u5728 0.01 \u4e4b\u9593\u3002\u5982 \u7b2c 3 \u7bc0\u63d0\u5230\u7684 LPP \u7684\u8a08\u7b97\u8907\u96dc\u5ea6\u9060\u4f4e\u65bc GOP\uff0c\u56e0\u6b64\u63a5\u4e0b\u4f86\u7684\u5be6\u9a57\u5c07\u4ee5 LPP \u4f5c\u70ba\u4e3b\u8981\u7684 \u767c\u97f3\u5206\u6578\u3002 \u8868", | |
| "type_str": "table", | |
| "content": "<table><tr><td>GOP GMM-HMM DNN-HMM \u63a5\u8457\u6211\u5011\u63a2\u8a0e\u767c\u97f3\u6aa2\u6e2c\u4efb\u52d9\u4f7f\u7528\u7b2c 3 \u7bc0\u6240\u63d0\u5230\u7684\u5c0d\u6578\u97f3\u7d20\u4e8b\u5f8c\u6a5f\u7387(LPP)\u4f5c\u70ba\u767c\u97f3 Correct pronunciation detection Mispronunciation Detection Recall Precision F1 Recall Precision F1 0.828 0.844 0.836 0.562 0.532 0.546 0.877 0.849 0.863 0.552 0.609 0.579 Mispronunciation Detection Recall Precision F1 Recall Precision F1 GOP 0.877 0.849 0.863 0.552 0.609 0.579 LPP 0.850 0.857 0.854 0.594 0.580 0.587 \u672c\u8ad6\u6587\u63a2\u8a0e\u7684\u4e3b\u984c\u70ba\u6700\u5927\u5316\u767c\u97f3\u6aa2\u6e2c\u6548\u80fd\u4e4b\u9451\u5225\u5f0f\u8a13\u7df4\u3002\u9996\u5148\u6211\u5011\u5b9a\u7fa9\u6b32\u9032\u884c\u767c\u97f3 \u6aa2\u6e2c\u7684\u8072\u5b78\u6a21\u578b\u8207\u6c7a\u7b56\u51fd\u6578\uff0c\u5ef6\u7e8c 5.2 \u5c0f\u7bc0\u7684\u8a9e\u97f3\u8fa8\u8b58\u57fa\u790e\u5be6\u9a57\u4e2d\u8868\u73fe\u6700\u597d\u7684\u8072\u5b78\u6a21\u578b DNN-HMM\uff0c\u4ee5\u53ca\u5728\u767c\u97f3\u6aa2\u6e2c\u4efb\u52d9\u4e0a\u6548\u679c\u4e0d\u905c\u8272\u65bc GOP \u6240\u63d0\u4f9b\u7684\u767c\u97f3\u5206\u6578\uff0c\u4e5f\u5c31\u662f LPP \u767c\u97f3\u5206\u6578\u4f5c\u70ba\u767c\u97f3\u5206\u6578\uff1b\u4e26\u900f\u904e\u975e\u7dda\u6027\u7684\u908f\u8f2f\u51fd\u6578\u8f49\u63db\u70ba\u6c7a\u7b56\u503c\uff0c\u4f9b\u6211\u5011\u8a08\u7b97\u8a55\u4f30\u8a72\u6a21 \u578b\u7684\u8868\u73fe\u3002\u7136\u800c\uff0c\u5728\u7b2c 4 \u7bc0\u8a0e\u8ad6\u7684F \u5ea6\u91cf\u4e4b\u76ee\u6a19\u51fd\u6578\u70ba\u932f\u8aa4\u767c\u97f3\u6aa2\u6e2c\u7684F \u5ea6\u91cf\uff0c\u4f46\u662f\u5728 \u96fb\u8166\u8f14\u52a9\u767c\u97f3\u8a13\u7df4\u7b49\u4efb\u52d9\u4e2d\u6b63\u78ba\u767c\u97f3\u6aa2\u6e2c\u4e5f\u662f\u975e\u5e38\u91cd\u8981\u7684\u90e8\u5206\uff0c\u6211\u5011\u4e0d\u5e0c\u671b\u7cfb\u7d71\u5c0d\u65bc\u5b78 \u7fd2\u8005\u672c\u8eab\u6b63\u78ba\u7684\u767c\u97f3\u9020\u6210\u8aa4\u5224\u3002\u5ef6\u7e8c\u7b2c 4 \u7bc0\u7684\u5b9a\u7fa9\u7684F \u5ea6\u91cf\u76ee\u6a19\u51fd\u6578\uff0c\u4e26\u518d\u6b21\u5b9a\u7fa9\u932f\u8aa4/ \u6b63\u78ba\u767c\u97f3\u6aa2\u6e2c\u7684F \u5ea6\u91cf\u8fd1\u4f3c\u7b97\u6cd5\uff1a \u5206\u6578\uff0c\u5982\u8868 5\u3002GOP \u8207 Correct pronunciation detection \u2211 \u2211 , \u2022 , \u039e \u2211 \u2211 ,</td></tr></table>", | |
| "num": null | |
| }, | |
| "TABREF7": { | |
| "html": null, | |
| "text": "\u57fa\u65bc LPP \u6700\u5927\u5316 F1 \u5ea6\u91cf\u9451\u5225\u5f0f\u8a13\u7df4\u65bc\u4e0d\u540c\u8a2d\u5b9a\u7684\u767c\u97f3\u6aa2\u6e2c\u6548\u80fd", | |
| "type_str": "table", | |
| "content": "<table><tr><td colspan=\"6\">\u8a55\u4f30\u5c3a\u5ea6\u76f8\u95dc\u6700\u4f73\u5316\u65b9\u6cd5\u65bc\u83ef\u8a9e\u932f\u8aa4\u767c\u97f3\u6aa2\u6e2c\u4e4b\u7814\u7a76</td><td>\u8a31\u66dc\u9e92 \u7b49 67</td></tr><tr><td colspan=\"7\">\u53ea\u66f4\u65b0\u8072\u5b78\u6a21\u578b\u53c3\u6578(+MFC (AM))\u53ef\u4ee5\u5f97\u5230\u66f4\u597d\u7684\u6548\u679c\uff0c\u82e5\u540c\u6642\u66f4\u65b0\u5169\u968e\u6bb5\u7684\u53c3\u6578 (+MFC (Both))\u6548\u679c\u6700\u4f73\u3002\u5f9e\u5be6\u9a57\u7d50\u679c\u4e2d\u53ef\u4ee5\u767c\u73fe\u66f4\u65b0\u8072\u5b78\u6a21\u578b\u53c3\u6578\u6709\u8457\u6700\u5927\u5e45\u5ea6\u7684\u9032\u6b65\uff0c \u8868 6. [Table 6. Mispronunciation detection results achieved by using LPP features with/without MFC training.] \u7531\u6b64\u53ef\u898b\u521d\u59cb\u7684\u8072\u5b78\u6a21\u578b\u662f\u70ba\u8a9e\u97f3\u8fa8\u8b58\u4efb\u52d9\u6240\u8a2d\u8a08\uff0c\u82e5\u7d93\u904e\u8abf\u9069\u5247\u53ef\u4ee5\u5f97\u5230\u66f4\u597d\u7684\u6548 \u679c\u3002 Correct pronunciation detection Mispronunciation detection</td></tr><tr><td/><td>Recall</td><td>Precision</td><td>F1</td><td>Recall</td><td>Precision</td><td>F1</td></tr><tr><td>LPP</td><td colspan=\"4\">Correct pronunciation detection (\u767c\u5c55\u96c6) 0.850 0.857 0.854 0.594</td><td>0.580</td><td>0.587</td></tr><tr><td>+MFC (DF)</td><td>0.863</td><td>LPP+MFC(DF) 0.866</td><td>0.865</td><td>Baseline 0.617</td><td>0.611</td><td>0.614</td></tr><tr><td>90.0% +MFC (AM)</td><td>0.906</td><td>0.870</td><td>0.888</td><td>0.612</td><td>0.694</td><td>0.650</td></tr><tr><td>88.0% +MFC (Both)</td><td>0.907</td><td>0.871</td><td>0.889</td><td>0.613</td><td>0.697</td><td>0.652</td></tr><tr><td colspan=\"7\">84.0% 86.0% 6. \u7d50\u8ad6\u8207\u672a\u4f86\u5c55\u671b F1-Score \u672c\u8ad6\u6587\u8457\u91cd\u5728\u96fb\u8166\u8f14\u52a9\u767c\u97f3\u8a13\u7df4\u7684\u932f\u8aa4\u767c\u97f3\u6aa2\u6e2c\u4efb\u52d9\uff0c\u4e26\u4ee5\u6700\u4f73\u5316\u932f\u8aa4\u767c\u97f3\u6aa2\u6e2c\u6548\u80fd\u70ba</td></tr><tr><td colspan=\"7\">\u6700\u5f8c\u6211\u5011\u4f7f\u7528\u53c3\u6578\u03c6\u4f5c\u70ba\u6b63\u78ba\u767c\u97f3\u8207\u932f\u8aa4\u767c\u97f3\u7684F \u5ea6\u91cf\u4e4b\u7dda\u6027\u7d44\u5408\u4f5c\u70ba\u6700\u7d42\u7684\u76ee\u6a19 \u51fd\u6578\uff1a \u039e \u2022 \u039e 1 \u2022 \u039e (24) \u5f9e\u6211\u5011\u7684\u5be6\u9a57\u4e2d\u53ef\u4ee5\u89c0\u5bdf\u5230\u53c3\u6578\u03c6\u5c0d\u65bc\u7d50\u679c\u767c\u97f3\u6aa2\u6e2c\u767c\u5c55\u96c6\u7684\u5f71\u97ff\uff0c\u5982\u5716 2\uff0c\u4ee5\u6700\u5927\u5316F \u5ea6\u91cf\u8abf\u6574\u6c7a\u7b56\u51fd\u6578\u7684\u53c3\u6578(\u672a\u66f4\u65b0\u8072\u5b78\u6a21\u578b)\u3002\u5f9e\u5716\u4e2d\u53ef\u4ee5\u767c\u73fe\u53c3\u6578\u03c6\u5c0d\u65bc\u932f\u8aa4\u767c\u97f3\u6aa2\u6e2c \u7d50\u679c\u7684\u5f71\u97ff\u5341\u5206\u986f\u8457(\u5716 2 \u4e0b\u534a\u90e8)\uff0c\u7576\u03c6=0.8 \u6642\u5728\u932f\u8aa4\u767c\u97f3\u6aa2\u6e2c\u6709\u6700\u597d\u7684\u6548\u679c(F \u5ea6\u91cf\u70ba 0.566\uff0c\u57fa\u790e\u5be6\u9a57\u7684F \u5ea6\u91cf\u70ba 0.527)\u3002\u4f46\u662f\u5728\u6b63\u78ba\u767c\u97f3\u4e2d\u4e26\u975e\u03c6=0.8 \u70ba\u6548\u679c\u6700\u4f73\uff0c\u4f46\u53c3\u6578 \u03c6\u7684\u8abf\u6574\u5c0d\u65bc\u6b63\u78ba\u767c\u97f3\u6aa2\u6e2c\u6548\u80fd\u7684\u5f71\u97ff\u8f03\u5c0f\uff0c\u56e0\u6b64\u6211\u5011\u6311\u9078\u53c3\u6578\u03c6\u5247\u4ee5\u932f\u8aa4\u767c\u97f3\u6aa2\u6e2c\u4e4b \u6548\u80fd\u70ba\u512a\u5148\u8003\u91cf\u3002\u6709\u8da3\u7684\u662f\uff0c\u5728\u8a13\u7df4\u8cc7\u6599\u4e2d\u6b63\u78ba\u767c\u97f3\u8207\u932f\u8aa4\u767c\u97f3\u7684\u6bd4\u4f8b\u6b63\u597d\u63a5\u8fd1 0.8\uff0c\u9019 \u8868\u793a\u900f\u904e\u8abf\u6574\u53c3\u6578\u03c6\u7684\u65b9\u5f0f\u5de7\u5999\u7684\u8655\u7406\u4e86\u8cc7\u6599\u985e\u5225\u4e0d\u5e73\u8861\u7684\u554f\u984c\u3002 \u57fa\u65bc\u5716 2 \u7684\u5be6\u9a57\u7d50\u679c\uff0c\u6211\u5011\u5c07\u56fa\u5b9a\u03c6=0.8 \u4e26\u4f9d\u7e8c\u63a2\u8a0e\u6700\u5927\u5316\u767c\u97f3\u6aa2\u6e2c\u6548\u80fd\u4e4b\u9451\u5225\u5f0f \u5716 2. \u4e0d\u540c\u03c6\u5728\u767c\u5c55\u96c6\u7684\u767c\u97f3\u6aa2\u6e2c\u6548\u80fd 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 \u03c6 52.0% 54.0% 56.0% 58.0% F1-Score \u03c6 Mispronunciation detection (\u767c\u5c55\u96c6) LPP+MFC(DF) Baseline \u4e3b\u8ef8\u9032\u884c\u4e00\u7cfb\u5217\u7684\u5be6\u9a57\u3002\u57fa\u65bc\u904e\u53bb\u5b78\u8005\u7684\u7814\u7a76\uff0c\u6211\u5011\u8a8d\u70ba\u4ee5\u6700\u5927\u5316\u767c\u97f3\u6aa2\u6e2c\u4e4bF \u5ea6\u91cf\u70ba \u76ee\u6a19\u51fd\u6578\u9032\u884c\u6a21\u578b\u8a13\u7df4\u662f\u975e\u5e38\u6709\u6f5b\u529b\u7684\u3002\u56e0\u6b64\u6211\u5011\u5ef6\u4f38\u8a72\u4f5c\u6cd5\u81f3\u73fe\u4eca\u8a9e\u97f3\u8fa8\u8b58\u6a21\u7d44\u5341\u5206 \u71b1\u9580\u7684\u90e8\u4efd-\u6df1\u5c64\u985e\u795e\u7d93\u7db2\u8def\u8072\u5b78\u6a21\u578b\uff0c\u53d6\u4ee3\u50b3\u7d71\u7684\u9ad8\u65af\u6df7\u5408\u8072\u5b78\u6a21\u578b\u3002\u5f9e\u5be6\u9a57\u7d50\u679c\u53ef \u4ee5\u767c\u73fe\u4ee5\u6700\u5927\u5316F \u5ea6\u91cf\u70ba\u76ee\u6a19\u5c0d\u6c7a\u7b56\u51fd\u6578\u6216\u8072\u5b78\u6a21\u578b\u7684\u53c3\u6578\u9032\u884c\u8abf\u6574\uff0c\u751a\u81f3\u662f\u540c\u6642\u8abf\u6574\uff0c \u90fd\u53ef\u4ee5\u5728\u6548\u679c\u4e0a\u5f97\u5230\u63d0\u5347\uff1b\u5c24\u5176\u5c0d\u65bc\u8072\u5b78\u6a21\u578b\u53c3\u6578\u9032\u884c\u8abf\u6574\u7684\u9032\u6b65\u5e45\u5ea6\u4ee4\u4eba\u5370\u8c61\u6df1\u523b\u3002 \u4e14\u4ee5F \u5ea6\u91cf\u4f5c\u70ba\u76ee\u6a19\u9032\u884c\u8a13\u7df4\u5728\u4e0d\u540c\u7684\u8a55\u4f30\u65b9\u5f0f\u4e5f\u53ef\u4ee5\u5f97\u5230\u9032\u6b65\u3002\u5728\u672a\u4f86\u6211\u5011\u5e0c\u671b\u5f9e\u7279 \u5fb5\u8207\u6a21\u578b\u7b49\u5169\u500b\u9762\u5411\u4f86\u63a2\u8a0e\u5c0d\u65bc\u96fb\u8166\u8f14\u52a9\u767c\u97f3\u8a13\u7df4\u4efb\u52d9\u7684\u5f71\u97ff\u3002\u5728\u7279\u5fb5\u7684\u90e8\u5206\uff0c\u6211\u5011\u671f \u671b\u5f9e\u4e0d\u540c\u89d2\u5ea6\u4f86\u7372\u53d6\u8ddf\u767c\u97f3\u72c0\u6cc1\u9ad8\u76f8\u95dc\u6027\u7684\u7279\u5fb5\uff0c\u5176\u4e2d\u97fb\u5f8b\u7279\u5fb5\u975e\u5e38\u5177\u6709\u6f5b\u529b\uff1b\u5728\u6a21\u578b \u7684\u90e8\u5206\u9664\u4e86\u6301\u7e8c\u63a2\u8a0e\u66f4\u65b0\u7a4e\u7684\u8072\u5b78\u6a21\u578b\u5916\uff0c\u6211\u5011\u4e5f\u9810\u671f\u5c07\u8a9e\u97f3\u8fa8\u8b58\u6240\u4f7f\u7528\u7684\u8abf\u9069\u6280\u8853\u79fb \u8f49\u5230\u8a72\u4efb\u52d9\uff0c\u4f8b\u5982\u4e00\u4e9b\u975e\u76e3\u7763\u5f0f\u7684\u8a9e\u8005\u8abf\u9069\u6216\u662f\u91dd\u5c0d\u4e0d\u540c\u8a9e\u8a00\u9032\u884c\u6a21\u578b\u8abf\u9069\u7b49\u3002 \u81f4\u8b1d \u672c\u8ad6\u6587\u4e4b\u7814\u7a76\u627f\u8499\u6559\u80b2\u90e8-\u570b\u7acb\u81fa\u7063\u5e2b\u7bc4\u5927\u5b78\u9081\u5411\u9802\u5c16\u5927\u5b78\u8a08\u756b(104-2911-I-003-301)\u8207 [Figure 282.0% \u884c \u653f \u9662 \u79d1 \u6280 \u90e8 \u7814 \u7a76 \u8a08 \u756b (MOST 104-2221-E-003-018-MY3 \u548c MOST 50.0% 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 105-2221-E-003-018-MY3)\u4e4b\u7d93\u8cbb\u652f\u6301\uff0c\u8b39\u6b64\u81f4\u8b1d\u3002</td></tr><tr><td colspan=\"7\">\u8a13\u7df4\u5c0d\u4e0d\u540c\u968e\u6bb5\u7684\u53c3\u6578\u9032\u884c\u8abf\u6574\u6240\u5e36\u4f86\u7684\u5f71\u97ff\u3002\u5728\u8868 6 \u4e2d\u6211\u5011\u57fa\u65bc LPP \u6240\u7b97\u51fa\u7684\u767c\u97f3\u5206</td></tr><tr><td colspan=\"7\">\u6578\u900f\u904e\u6c7a\u7b56\u51fd\u6578\u4e26\u7b97\u51fa\u6574\u9ad4\u7684F \u5ea6\u91cf\uff0c\u4ee5\u6700\u5927\u5316F \u5ea6\u91cf\u66f4\u65b0\u6c7a\u7b56\u51fd\u6578(+MFC (DF))\u6216\u8072\u5b78</td></tr><tr><td colspan=\"7\">\u6a21\u578b(+MFC (AM))\u7684\u53c3\u6578\uff0c\u751a\u81f3\u662f\u6c7a\u7b56\u51fd\u6578\u8207\u8072\u5b78\u6a21\u578b\u7684\u53c3\u6578\u540c\u6642\u8abf\u6574(+MFC (Both))\u3002</td></tr><tr><td colspan=\"7\">\u5f9e\u8868 6 \u53ef\u4ee5\u767c\u73fe\u7121\u8ad6\u662f\u66f4\u65b0\u4efb\u4f55\u968e\u6bb5\u7684\u53c3\u6578\u5728\u767c\u97f3\u6aa2\u6e2c\u4efb\u52d9\u4e0a\u90fd\u53ef\u4ee5\u5f97\u5230\u986f\u8457\u7684\u63d0\u5347\u3002</td></tr><tr><td colspan=\"7\">\u9996\u5148\u8a0e\u8ad6\u66f4\u65b0\u6c7a\u7b56\u51fd\u6578\u7684\u53c3\u6578(+MFC (DF))\uff0c\u8207\u57fa\u790e\u5be6\u9a57\u76f8\u6bd4(LPP)\u5247\u6709\u660e\u986f\u7684\u63d0\u5347\uff1b\u800c</td></tr></table>", | |
| "num": null | |
| } | |
| } | |
| } | |
| } |