| { |
| "paper_id": "O04-1002", |
| "header": { |
| "generated_with": "S2ORC 1.0.0", |
| "date_generated": "2023-01-19T08:01:00.132999Z" |
| }, |
| "title": "\u805a\u96c6\u4e8b\u5f8c\u6a5f\u7387\u7dda\u6027\u8ff4\u6b78\u8abf\u9069\u6f14\u7b97\u6cd5\u61c9\u7528\u65bc\u8a9e\u97f3\u8fa8\u8b58 Aggregate a Posteriori Linear Regression for Speech Recognition", |
| "authors": [], |
| "year": "", |
| "venue": null, |
| "identifiers": {}, |
| "abstract": "", |
| "pdf_parse": { |
| "paper_id": "O04-1002", |
| "_pdf_hash": "", |
| "abstract": [], |
| "body_text": [ |
| { |
| "text": "\u3001learning vector quantization(LVQ) [18] \uff0c\u5230\u8fd1\u4f86\u7684\u6700\u5c0f\u5206\u985e\u932f\u8aa4(minimum classification error, MCE) [11] \u3001\u6700\u5927\u76f8\u4e92\u8cc7\u8a0a(maximum mutual information, MMI) [20] \uff0c\u6709\u8a31\u591a\u4e0d\u540c\u7684\u7406\u8ad6\u65b9\u6cd5\u3002\u9451\u5225", |
| "cite_spans": [ |
| { |
| "start": 35, |
| "end": 39, |
| "text": "[18]", |
| "ref_id": "BIBREF17" |
| }, |
| { |
| "start": 87, |
| "end": 91, |
| "text": "[11]", |
| "ref_id": "BIBREF10" |
| }, |
| { |
| "start": 133, |
| "end": 137, |
| "text": "[20]", |
| "ref_id": "BIBREF19" |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "\u5f0f \u8a13 \u7df4 \u8207 \u5176 \u5b83 \u6a21 \u578b \u8a13 \u7df4 \u65b9 \u6cd5 \u6700 \u5927 \u7684 \u4e0d \u540c \u662f \uff0c \u9664 \u4e86 \u8003 \u616e \u6a23 \u672c \u8207 \u672c \u8eab \u6a21 \u578b \u7684 \u76f8 \u4f3c \u5ea6 \u4e4b \u5916 \uff0c \u9084 \u984d \u5916 \u8003 \u616e \u6a23 \u672c \u8207 \u5176 \u5b83 \u6a21 \u578b \u4e4b \u9593 \u7684 \u76f8 \u4f3c \u5ea6 \uff0c \u9019 \u7a2e \u4f5c \u6cd5 \u53ef \u4ee5 \u907f \u514d \u6a21 \u578b \u8a13 \u7df4 \u6642 \uff0c \u539f \u672c \u5c31 \u76f8 \u4f3c \u7684 \u8a9e \u97f3 \u6a21 \u578b \u7522 \u751f \u4e92 \u76f8 \u6df7 \u6dc6 \u7684 \u60c5 \u6cc1 \u3002 Qi Li [15]\u5728 2002 \u5e74 \u63d0 \u51fa \u4e00 \u822c \u5316 \u6700 \u5c0f \u932f \u8aa4 \u7387 (generalized minimum error rate, GMER)\uff0c\u7531\u4e8b\u5f8c\u6a5f\u7387\u7684\u89d2\u5ea6\u51fa\u767c\uff0c \u5b9a \u7fa9 \u805a \u96c6 \u4e8b \u5f8c \u6a5f \u7387 (aggregate a posteriori, AAP)\uff0c\u4e26\u5c07\u4e8b\u5f8c\u6a5f\u7387\u6539\u5beb\u70ba\u5177\u9451\u5225\u6027\u5f62\u5f0f\u7684\u8aa4\u8fa8\u7387(misclassification measure) \u51fd \u5f0f \u3002 \u5728 \u8a13 \u7df4 \u6a21 \u578b \u53c3 \u6578 \u4e0a \uff0c \u4e0d \u4f7f \u7528 \u4e00 \u822c \u7684 \u5ee3 \u7fa9 \u6a5f \u7387 \u905e \u6e1b \u6cd5 \u5247 (generalized probabilistic descent, GPD)\uff0c\u900f\u904e\u4e00\u4e9b\u689d\u4ef6\u5047 \u8a2d \uff0c \u5373 \u53ef \u63a8 \u5c0e \u51fa \u6a21 \u578b \u53c3 \u6578 \u4f30 \u6e2c \u7684 \u5c01 \u9589 \u89e3 \u5f62 \u5f0f \u3002 \u5728 \u8a9e \u8005 \u8abf \u9069 \u7684 \u7814 \u7a76 \u4e0a \uff0c \u6700 \u5ee3 \u70ba \u4f7f \u7528 \u7684 \u6709 \u6700 \u5927 \u76f8 \u4f3c \u5ea6 \u7dda \u6027 \u8ff4 \u6b78 (maximum likelihood linear regression, MLLR)\u8abf\u9069 [7][14]\u8207\u6700\u5927\u4e8b\u5f8c\u6a5f\u7387\u8abf\u9069\u5169\u5927\u985e\u65b9\u6cd5\u3002\u5728\u672c\u7814\u7a76\u4e2d\u6211\u5011\u5c07\u4f7f\u7528\u524d\u8005\u4f5c\u70ba\u8abf\u9069\u7684\u4e3b\u8981\u67b6\u69cb\uff0c\u900f\u904e\u6240\u4f30\u6e2c\u51fa\u4e4b\u7dda\u6027\u8ff4 \u6b78 \u77e9 \u9663 \u5c0d \u8a9e \u97f3 \u6a21 \u578b \u53c3 \u6578 \u9032 \u884c \u8abf \u9069 \u3002 \u7531 \u65bc \u8003 \u616e \u5230 \u4f7f \u7528 \u8a9e \u6599 \u91cf \u7a00 \u5c11 \u6613 \u9020 \u6210 \u8abf \u9069 \u6548 \u679c \u5931 \u6e96 \u7684 \u60c5 \u6cc1 \uff0c \u5f15 \u5165 \u7dda \u6027 \u8f49 \u63db \u77e9 \u9663 \u4e4b \u4e8b \u524d \u5206 \u4f48 \u8cc7 \u8a0a \uff0c \u4ee5 \u5f37 \u5065 \u5316 \u8abf \u9069 \u6548 \u80fd \u5916 \uff0c \u4e5f \u5c07 \u7531 \u9451 \u5225 \u5f0f \u8a13 \u7df4 \u4e4b \u89d2 \u5ea6 \u51fa \u767c \uff0c \u5617 \u8a66 \u627e \u51fa \u4e0d \u540c \u65bc \u50b3 \u7d71 \u4ee5 \u8c9d \u6c0f \u6cd5 \u5247 \u70ba \u6e96 \u4e4b \u6700 \u5927 \u5316 \u805a \u96c6 \u4e8b \u5f8c \u6a5f \u7387 \u7dda \u6027 \u8ff4 \u6b78", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "(aggregate a posteriori linear regression, AAPLR)\u6f14\u7b97\u6cd5\u3002\u6545\u6211\u5011\u6703\u91dd\u5c0d\u6587\u737b\u4e2d\u6240\u63d0\u904e\u4e4b\u4ee5\u7dda\u6027 \u8ff4 \u6b78 \u70ba \u4e3b \u4e4b \u8abf \u9069 \u6f14 \u7b97 \u6cd5 \u4f5c \u56de \u9867 \u3002 \u9664 \u4e86 \u6700 \u5927 \u76f8 \u4f3c \u5ea6 \u7dda \u6027 \u8ff4 \u6b78 \u8abf \u9069 \u6f14 \u7b97 \u6cd5 \u4e4b \u5916 \uff0c \u4e3b \u8981 \u6709 \u6700 \u5927 \u4e8b \u5f8c \u6a5f \u7387 \u7dda \u6027 \u8207 \u6240 \u4f30 \u6e2c \u985e \u5225 \u53c3 \u6578 \u9593 \u4e4b \u9451 \u5225 \u6027 \u6cd5 \u5247 \u800c \u5f97 \u5230 \u8f03 \u50b3 \u7d71 \u6700 \u5927 \u76f8 \u4f3c \u5ea6 \u4f30 \u6e2c \u66f4 \u597d \u7684 \u5206 \u985e \u6b63 \u78ba \u7387 \uff0c \u53c8 \u5f9e \u63a8 \u5c0e \u6700 \u7d42 \u7d50 \u679c \u4e4b \u5c01 \u9589 \u89e3 \u800c \u53ef \u7372 \u5f97 \u5feb \u901f \u8abf \u9069 \u7684 \u6548 \u80fd \u3002 \u5728 \u5be6 \u9a57 \u4e2d \uff0c \u6211 \u5011 \u53ef \u4ee5 \u770b \u5230 \u7121 \u8ad6 \u5728 \u4efb \u4f55 \u8abf \u9069 \u8cc7 \u6599 \u91cf \u4e4b \u4e0b \uff0c \u6240 \u63d0 \u51fa \u4e4b \u8abf \u9069 \u6f14 \u7b97 \u6cd5 \u4e4b \u6548 \u80fd \u53ef \u4ee5 \u6bd4 \u5176 \u4ed6 \u540c \u6a23 \u57fa \u65bc \u7dda \u6027 \u8ff4 \u6b78 \u8abf \u9069 \u70ba \u4e3b \u4e4b \u6f14 \u7b97 \u6cd5 \u6709 \u66f4 \u597d \u7684 \u6548 \u80fd \u8868 \u73fe \u3002 \u5728 \u672c \u8ad6 \u6587 \u4e2d \uff0c \u4e00 \u822c \u5316 \u6700 \u5c0f \u932f \u8aa4 \u7387 \u4e2d \u4e4b \u985e \u5225 \u6a5f \u7387 \u4ee5 \u4e00 \u5e38 \u6578 \u8868 \u793a \uff0c \u5728 \u6a21 \u578b \u53c3 \u6578 \u4f30 \u6e2c \u4e2d \u8f03 \u4e0d \u5177 \u53c3 \u8003 \u50f9 \u503c \uff0c \u6216 \u8a31 \u5617 \u8a66 ", |
| "cite_spans": [], |
| "ref_spans": [ |
| { |
| "start": 70, |
| "end": 200, |
| "text": "\u8ff4 \u6b78 \u70ba \u4e3b \u4e4b \u8abf \u9069 \u6f14 \u7b97 \u6cd5 \u4f5c \u56de \u9867 \u3002 \u9664 \u4e86 \u6700 \u5927 \u76f8 \u4f3c \u5ea6 \u7dda \u6027 \u8ff4 \u6b78 \u8abf \u9069 \u6f14 \u7b97 \u6cd5 \u4e4b \u5916 \uff0c \u4e3b \u8981 \u6709 \u6700 \u5927 \u4e8b \u5f8c \u6a5f \u7387 \u7dda \u6027", |
| "ref_id": null |
| }, |
| { |
| "start": 201, |
| "end": 721, |
| "text": "\u8207 \u6240 \u4f30 \u6e2c \u985e \u5225 \u53c3 \u6578 \u9593 \u4e4b \u9451 \u5225 \u6027 \u6cd5 \u5247 \u800c \u5f97 \u5230 \u8f03 \u50b3 \u7d71 \u6700 \u5927 \u76f8 \u4f3c \u5ea6 \u4f30 \u6e2c \u66f4 \u597d \u7684 \u5206 \u985e \u6b63 \u78ba \u7387 \uff0c \u53c8 \u5f9e \u63a8 \u5c0e \u6700 \u7d42 \u7d50 \u679c \u4e4b \u5c01 \u9589 \u89e3 \u800c \u53ef \u7372 \u5f97 \u5feb \u901f \u8abf \u9069 \u7684 \u6548 \u80fd \u3002 \u5728 \u5be6 \u9a57 \u4e2d \uff0c \u6211 \u5011 \u53ef \u4ee5 \u770b \u5230 \u7121 \u8ad6 \u5728 \u4efb \u4f55 \u8abf \u9069 \u8cc7 \u6599 \u91cf \u4e4b \u4e0b \uff0c \u6240 \u63d0 \u51fa \u4e4b \u8abf \u9069 \u6f14 \u7b97 \u6cd5 \u4e4b \u6548 \u80fd \u53ef \u4ee5 \u6bd4 \u5176 \u4ed6 \u540c \u6a23 \u57fa \u65bc \u7dda \u6027 \u8ff4 \u6b78 \u8abf \u9069 \u70ba \u4e3b \u4e4b \u6f14 \u7b97 \u6cd5 \u6709 \u66f4 \u597d \u7684 \u6548 \u80fd \u8868 \u73fe \u3002 \u5728 \u672c \u8ad6 \u6587 \u4e2d \uff0c \u4e00 \u822c \u5316 \u6700 \u5c0f \u932f \u8aa4 \u7387 \u4e2d \u4e4b \u985e \u5225 \u6a5f \u7387 \u4ee5 \u4e00 \u5e38 \u6578 \u8868 \u793a \uff0c \u5728 \u6a21 \u578b \u53c3 \u6578 \u4f30 \u6e2c \u4e2d \u8f03 \u4e0d \u5177 \u53c3 \u8003 \u50f9 \u503c \uff0c \u6216 \u8a31 \u5617 \u8a66", |
| "ref_id": null |
| } |
| ], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "\u6211 \u5011 \u5c07 \u63d0 \u51fa \u7684 \u8a9e \u97f3 \u6a21 \u578b \u53c3 \u6578 \u8abf \u9069 \u6f14 \u7b97 \u6cd5 \uff0c \u4f7f \u7528 \u9023 \u7e8c \u8a9e \u97f3 \u8fa8 \u8b58 \u7cfb \u7d71 \u9032 \u884c \u8207 \u5176 \u4ed6 \u8abf \u9069 \u6f14 \u7b97 \u6cd5 \u7684 \u6548 \u80fd \u8a55 \u4f30 \u3002 \u63a5 \u4e0b \u4f86 \uff0c \u6211 \u5011 \u5148 \u56de \u9867 \u8fd1 \u5e74 \u4f86 \u9451 \u5225 \u5f0f \u8a13 \u7df4 \u7684 \u76f8 \u95dc \u7814 \u7a76 \u6587 \u737b \u8207 \u524d \u8ff0 \u5e7e \u7a2e \u4ee5 \u8f49 \u63db \u70ba \u4e3b \u4e4b \u8a9e \u97f3 \u6a21 \u578b \u8abf \u9069 \u6f14 \u7b97 \u6cd5 \u53ca \u5c07 \u805a \u96c6 \u4e8b \u5f8c \u6a5f \u7387 \u61c9 \u7528 \u5728 \u9451 \u5225 \u5f0f \u8072 \u5b78 \u6a21 \u578b \u53c3 \u6578 \u4f30 \u6e2c \u4e0a \u7684 \u65b9 \u6cd5 \u3002 \u5176 \u6b21 \uff0c \u8aaa \u660e \u6211 \u5011 \u5c07 \u4e00 \u822c \u5316 \u6700 \u5c0f \u932f \u8aa4 \u7387 \u61c9 \u7528 \u5728 \u8a9e \u97f3 \u6a21 \u578b \u53c3 \u6578 \u8abf \u9069 \u53ca \u5728 \u8abf \u9069 \u6642 \u8003 \u616e \u8f49 \u63db \u77e9 \u9663 \u7684 \u4e8b \u524d \u6a5f \u7387 \u5206 \u4f48 \uff0c \u6700 \u5f8c \u5f97 \u5230 \u4f30 \u6e2c \u7684 \u8f49 \u63db \u77e9 \u9663 \u53c3 \u6578 \u5c01 \u9589 \u89e3 \u4e4b \u76f8 \u95dc \u7406 \u8ad6 \u5167 \u5bb9 \u3002 \u63a5 \u8457 \u8aaa \u660e \u5be6 \u9a57 \u8a2d \u5b9a \u8207 \u9032 \u884c \u65b9 \u5f0f \u4e26 \u7531 \u5be6 \u9a57 \u6240 \u5f97 \u7d50 \u679c \u9032 \u884c \u8a0e \u8ad6 \u3002 \u5728 \u7d50 \u5c3e \u90e8 \u4efd \u5247 \u7c21 \u55ae \u6b78 \u7d0d \u672c \u8ad6 \u6587 \u7684 \u4e3b \u8981 \u91cd \u9ede \u8207 \u7d50 \u8ad6 \uff0c \u4e26 \u8aaa \u660e \u672a \u4f86 \u7e7c \u7e8c \u7814 \u7a76 \u7684 \u65b9 \u5411 \u8207 \u8ab2 \u984c \u3002 2. \u9451\u5225\u5f0f\u8a13\u7df4\u53ca\u7dda\u6027\u56de\u6b78\u8abf\u6574 \u6700 \u5927 \u76f8 \u4f3c \u5ea6 \u53c3 \u6578 \u4f30 \u6e2c \u6cd5 \u5247 \u662f \u6700 \u666e \u904d \u7528 \u4f86 \u8a13 \u7df4 \u96b1 \u85cf \u5f0f \u99ac \u53ef \u592b \u6a21 \u578b \u53c3 \u6578 \u7684 \u65b9 \u6cd5 \uff0c \u5b83 \u5229 \u7528 EM \u6f14 \u7b97 \u6cd5 \u4f30 \u6e2c \u6a21 \u578b \u53c3 \u6578 \u975e \u5e38 \u6709 \u6548 \u7387 \uff1b \u6700 \u5927 \u76f8 \u4f3c \u5ea6 \u7684 \u7f3a \u9ede \u662f \u6a21 \u578b \u53c3 \u6578 \u53ea \u5229 \u7528 \u5c6c \u65bc \u672c \u8eab \u6a21 \u578b \u7684 \u8cc7 \u6599 \u4f86 \u4f30 \u6e2c \uff0c \u548c \u5176 \u5b83 \u6a21 \u578b \u7684 \u53c3 \u6578 \u4f30 \u6e2c \u57fa \u672c \u4e0a \u662f \u7368 \u7acb \u7684 \u3002 \u6700 \u5c0f \u5206 \u985e \u932f \u8aa4 \u548c \u6700 \u5927 \u4ea4 \u4e92 \u8cc7 \u8a0a \uff0c \u662f \u8fd1 \u4f86 \u8f03 \u5ee3 \u70ba \u5229 \u7528 \u7684 \u9451 \u5225 \u5f0f \u8a13 \u7df4 \u65b9 \u6cd5 \uff0c \u9664 \u4e86 \u8a13 \u7df4 \u8a9e \u97f3 \u6a21 \u578b \u5916 \uff0c \u9084 \u7528 \u5728 \u8a9e \u8a00 \u6a21 \u578b (language model)\u7684\u8a13\u7df4\u4e0a[13]\u3001\u8a9e\u8005\u8fa8\u8b58\u6a21\u578b\u8a13\u7df4\u3001\u7279\u5fb5\u53c3\u6578\u64f7\u53d6\u3002\u4f7f\u7528\u9451\u5225\u5f0f\u8a13\u7df4\u4f30\u6e2c\u6a21\u578b\u53c3\u6578\u6642\uff0c\u9664\u4e86\u672c\u8eab \u6a21 \u578b \u7684 \u8cc7 \u6599 \u5916 \uff0c \u9084 \u8003 \u616e \u8207 \u5176 \u5b83 \u6a21 \u578b \u53c3 \u6578 \u4e4b \u9451 \u5225 \u6027 \uff0c \u6240 \u4ee5 \u53ef \u4ee5 \u66f4 \u6b63 \u78ba \u5730 \u4f30 \u6e2c \u51fa \u6240 \u9700 \u7684 \u6a21 \u578b \u53c3 \u6578 \u5167 \u5bb9 \u3002 \u5728 [15][16]\u4e2d\uff0c\u4f5c \u8005 \u63d0 \u51fa \u4e86 \u53e6 \u4e00 \u7a2e \u9451 \u5225 \u5f0f \u8a13 \u7df4 \u65b9 \u6cd5 \uff0c \u7a31 \u4f5c \u4e00 \u822c \u5316 \u6700 \u5c0f \u932f \u8aa4 \u7387 \uff0c \u5f9e \u4e8b \u5f8c \u6a5f \u7387 \u51fa \u767c \uff0c \u5b9a \u7fa9 \u8207 \u6700 \u5927 \u4e8b \u5f8c \u6a5f \u7387 \u76f8 \u4f3c \u7684 \u76ee \u6a19 \u51fd \u5f0f \uff0c \u4e26 \u4e14 \u6539 \u5beb \u70ba \u9451 \u5225 \u5f0f \u8a13 \u7df4 \u7684 \u5f62 \u5f0f \uff0c \u4ee5 \u4e0b \u5206 \u5225 \u7c21 \u4ecb \u9019 \u4e09 \u7a2e \u9451 \u5225 \u5f0f \u8a13 \u7df4 \u6cd5 \u5247 \u3002 2.1 \u6700 \u5c0f \u5206 \u985e \u932f \u8aa4 (MCE)\u8a13\u7df4\u6cd5\u5247 \u5728 \u5169 \u500b \u985e \u5225 2 1 ,C C \u7684 \u5206 \u985e \u5668 \u88e1 \uff0c \u5047 \u8a2d 1 C \u2208 x \uff0c \u8c9d \u6c0f \u5206 \u985e \u6cd5 \u5247 \u5b9a \u7fa9 \u4e86 \u6700 \u57fa \u672c \u7684 \u8aa4 \u8fa8 \u503c \u51fd \u5f0f (misclassification measure)\u70ba ) | ( ) | ( ) ( 1 2 x x x C P C P d \u2212 = (1) \u4e0a \u5f0f \u8868 \u793a \u985e \u5225 1 C \u7684 \u89c0 \u5bdf \u8cc7 \u6599 x \u88ab \u5206 \u985e \u5668 \u5206 \u985e \u5230 \u985e \u5225 2 C \u7684 \u53ef \u80fd \u6027 \uff0c \u5728 \u591a \u500b \u985e \u5225 \u7684 \u5206 \u985e \u5668 [12]\u88e1\uff0c\u5b9a\u7fa9\u8aa4\u8fa8\u503c\u51fd\u5f0f [ ] \u2211 \u2208 \u039b \u2212 \u039b = i M i k i k k g g m d ) ; ( ) ; ( 1 ) ( x x x (2) \u5176 \u4e2d ) ; ( \u039b x i g \u70ba \u89c0 \u5bdf \u8cc7 \u6599 x \u5c0d \u985e \u5225 i C \u7684 \u76f8 \u4f3c \u5ea6 \uff0c \u039b \u8868 \u793a \u6240 \u6709 \u985e \u5225 \u7684 \u6a21 \u578b \u53c3 \u6578 \uff0c { } ) ; ( ) ; ( | \u039b > \u039b = x x k i k g g i M \uff0c \u4ee3 \u8868 \u4e00 \u7fa4 \u5c0d \u89c0 \u5bdf \u8cc7 \u6599 x \u7684 \u76f8 \u4f3c \u5ea6 \u6bd4 \u985e \u5225 k C \u5c0d \u89c0 \u5bdf \u8cc7 \u6599 x \u76f8 \u4f3c \u5ea6 \u66f4 \u5177 \u7af6 \u722d \u6027 \u7684 \u985e \u5225 \u96c6 \u5408 \uff0c \u5373 \u6df7 \u6dc6 \u985e \u5225 (confusing classes)\u6216\u7af6\u722d\u985e\u5225(competing classes)\u7684\u96c6\u5408\u3002 \u5f0f \u5b50 (2)\u4e2d\uff0c k S \u4e26 \u975e \u662f \u56fa \u5b9a \u7684 \u96c6 \u5408 \uff0c \u5b83 \u96a8 \u8457 \u6a21 \u578b \u53c3 \u6578 \u039b \u548c \u89c0 \u5bdf \u8cc7 \u6599 x \u800c \u6539 \u8b8a \uff0c \u800c \u4e14 \u8a72 \u5f0f \u5728 \u039b \u4e0d \u9023 \u7e8c [12]\uff0c\u9019 \u5728 \u6700 \u9661 \u5761 \u964d \u6cd5 (gradient descent)\u88e1\u4e26\u4e0d\u9069\u7528\uff0c\u56e0\u6b64\u53e6\u5916\u5b9a\u7fa9\u4e86\u4e00\u500b\u9023\u7e8c\u6027\u7684\u8aa4\u8fa8\u503c\u516c\u5f0f\u70ba \u03b7 \u03b7 / 1 , ) ; ( 1 1 ) ; ( ) ( \u23a5 \u23a6 \u23a4 \u23a2 \u23a3 \u23a1 \u039b \u2212 + \u039b \u2212 = \u2211 \u2260k j j j k k g M g d x x x (3) \u5176 \u4e2d \u03b7 \u662f \u4e00 \u500b \u6b63 \u6578 \uff0c \u85c9 \u8457 \u6539 \u8b8a \u03b7 \u7684 \u503c \uff0c \u53ef \u4ee5 \u6539 \u8b8a \u5f0f \u5b50 \u88e1 \u5177 \u5f71 \u97ff \u529b \u7684 \u7af6 \u722d \u985e \u5225 \u6578 \u91cf \uff0c \u4ee4 \u221e \u2192 \u03b7 \uff0c \u4e00 \u500b \u6975 \u7aef \u7684 \u8aa4 \u8fa8 \u503c \u516c \u5f0f \u70ba ) ; ( ) ; ( ) ( \u039b + \u039b \u2212 = x x x i k k g g d (4) \u985e \u5225 i C \u662f \u9664 \u4e86 \u985e \u5225 k C \u5916 \uff0c \u548c \u89c0 \u5bdf \u8cc7 \u6599 x \u76f8 \u4f3c \u5ea6 \u6700 \u5927 \u7684 \u985e \u5225 \uff0c 0 ) ( > x k d \u4ee3 \u8868 \u767c \u751f \u5206 \u985e \u932f \u8aa4 \uff0c 0 ) ( \u2264 x k d \u4ee3 \u8868 \u6b63 \u78ba \u5206 \u985e \u3002 \u70ba \u4e86 \u66f4 \u9032 \u4e00 \u6b65 \u5b8c \u6210 \u76ee \u6a19 \u51fd \u5f0f \u7684 \u5b9a \u7fa9 \uff0c \u628a \u8aa4 \u8fa8 \u503c \u516c \u5f0f \u4ee3 \u5165 cost function )) ( ( ) ; ( x k k d l x l = \u039b (5) cost function \u4e00 \u822c \u70ba \u9023 \u7e8c \u6027 \uff0c \u7bc4 \u570d \u70ba [0,1]\u7684\u51fd\u5f0f\uff0c\u6700\u5e38\u7528\u65bc MCE \u7684 \u70ba sigmoid\uff0c ) exp( 1 1 ) ( \u03b8 \u03b3 + \u2212 + = k k d d l (6) \u5c0d \u65bc \u67d0 \u500b \u89c0 \u5bdf \u8cc7 \u6599 x \uff0c \u6211 \u5011 \u53ef \u4ee5 cost function \u5b9a \u7fa9 \u5206 \u985e \u5668 \u7684 \u6548 \u7387 \u70ba \u2211 = \u2208 \u039b = \u039b M i i i C l l 1 ) ( 1 ) ; ( ) ; ( x x x (7) \u6700 \u5f8c \u5229 \u7528 \u5ee3 \u7fa9 \u6a5f \u7387 \u905e \u6e1b (generalized probabilistic decent, GPD)\u6f14\u7b97\u6cd5\u9032\u884c\u758a\u4ee3\u904b\u7b97\u4ee5\u5be6\u73fe MCE \u6cd5 \u5247 \u3002 t x l U t t t t \u039b = \u039b + \u039b \u2207 \u2212 \u039b = \u039b | ) ; ( 1 \u03b5 (8) \u5ee3 \u7fa9 \u6a5f \u7387 \u905e \u6e1b \u6cd5 \u5247 \u662f \u61c9 \u7528 \u5f88 \u5ee3 \u7684 \u6f14 \u7b97 \u6cd5 \uff0c \u5229 \u7528 \u53cd \u8986 \u7684 \u8a08 \u7b97 \uff0c \u905e \u8ff4 \u5f97 \u5230 \u4e00 \u6536 \u6582 \u7684 \u503c \uff0c \u7f3a \u9ede \u662f \u6536 \u6582 \u901f \u5ea6 \u6162 \uff0c \u800c \u4e14 \u5f0f \u4e2d \u7684 \u5b78 \u7fd2 \u4fc2 \u6578 t \u03b5 \u9700 \u5c0d \u61c9 \u4e0d \u540c \u7684 \u8cc7 \u6599 \u7279 \u6027 \u53bb \u8abf \u6574 \u3002 \u66f4 \u9032 \u4e00 \u6b65 \u4e4b \u76f8 \u95dc \u53c3 \u6578 \u4f30 \u6e2c \u904e \u7a0b \u8207 \u7d50 \u679c \u8a73 \u898b [11]\u3002 2.2 \u6700 \u5927 \u4ea4 \u4e92 \u8cc7 \u8a0a (MMI)\u8a13\u7df4\u6cd5\u5247 \u9664 \u4e86 \u6700 \u5c0f \u5206 \u985e \u932f \u8aa4 \u6cd5 \u5247 \u5916 \uff0c \u6700 \u5927 \u4ea4 \u4e92 \u8cc7 \u8a0a \u4e5f \u662f \u666e \u904d \u5229 \u7528 \u7684 \u9451 \u5225 \u5f0f \u8a13 \u7df4 \u5f0f \u6cd5 \u5247 [1][20]\uff0c\u6700\u5927\u4ea4\u4e92\u8cc7\u8a0a\u8f03\u96b1\u6027\u7684\u5f15 \u5165 \u4e86 \u89c0 \u5bdf \u8cc7 \u6599 \u8207 \u5176 \u5b83 \u985e \u5225 \u7684 \u76f8 \u4f3c \u5ea6 \uff0c \u6240 \u4ee5 \u8207 \u4e00 \u822c \u5316 \u6700 \u5c0f \u932f \u8aa4 \u7387 \u8f03 \u76f8 \u4f3c \uff0c \u5728 \u6df7 \u5408 \u6578 \u9ad8 \u7684 \u60c5 \u6cc1 \u4e0b \uff0c \u6700 \u5927 \u4ea4 \u4e92 \u8cc7 \u8a0a \u80fd \u8a13 \u7df4 \u51fa \u6bd4 \u6700 \u5c0f \u5206 \u985e \u932f \u8aa4 \u8fa8 \u8b58 \u7387 \u66f4 \u9ad8 \u7684 \u6a21 \u578b \u53c3 \u6578 [1]\uff0c\u7531\u65bc\u6700\u5927\u4ea4\u4e92\u8cc7\u8a0a\u8003\u616e\u4e86\u89c0\u5bdf\u8cc7\u6599\u548c\u6240\u6709\u985e\u5225\u7684\u76f8\u4f3c\u5ea6\uff0c\u56e0\u6b64\u6bd4\u6700\u5c0f \u5206 \u985e \u932f \u8aa4 \u5728 \u5be6 \u4f5c \u4e0a \u96e3 \u5ea6 \u66f4 \u9ad8 \u3002 \u70ba \u4e86 \u5feb \u901f \u8a08 \u7b97 \u96b1 \u85cf \u5f0f \u99ac \u53ef \u592b \u6a21 \u578b \u548c \u89c0 \u5bdf \u8cc7 \u6599 x \u7684 \u76f8 \u4f3c \u5ea6 \uff0c \u5fc5 \u9808 \u4f7f \u7528 forward-backward \u6f14 \u7b97 \u6cd5 \u3002 \u900f \u904e forward probability ) (t j \u03b1 \u8207 backward probability ) (t j \u03b2 \u7684 \u8868 \u793a \uff0c \u985e \u5225 m C \u7522 \u751f \u89c0 \u5bdf \u8cc7 \u6599 X \u7684 \u6a5f \u7387 \u53ef \u5beb \u70ba \u4e0b \u5f0f \u2211\u2211 = = = \u039b T t N j j j m t t C X P 1 1 ) ( ) ( ) , | ( \u03b2 \u03b1 (9) \u5b9a \u7fa9 \u985e \u5225 m C \u8207 \u89c0 \u5bdf \u8cc7 \u6599 X \u7684 \u4ea4 \u4e92 \u8cc7 \u8a0a \u70ba \u2211 = \u039b \u2212 = \u2212 = = M m m m m m m m C P C P C P P C P P C P C I 1 ' ' ' ) ( ) | ( log ) | ( log ) ( log ) | ( log ) ( ) | ( log ) , ( X X X X X X X (10) \u5176 \u4e2d ) , ( m C P X \u4ee3 \u8868 \u985e \u5225 m C \u8207 X \u540c \u6642 \u51fa \u73fe \u7684 \u6a5f \u7387 \uff0c \u5373 \u806f \u5408 \u76f8 \u4f3c \u5ea6 (joint likelihood)\u3002\u7531(10)\u5f0f\u53ef\u770b\u51fa\uff0c\u9664\u4e86\u89c0\u5bdf\u8cc7 \u6599 X \u8207 \u5c0d \u61c9 \u985e \u5225 m C \u7684 \u76f8 \u4f3c \u5ea6 \u4e4b \u5916 \uff0c \u9084 \u52a0 \u5165 \u4e86 X \u8207 \u5176 \u5b83 \u985e \u5225 \u7684 \u76f8 \u4f3c \u5ea6 \u4f5c \u70ba \u53c3 \u6578 \u4f30 \u6e2c \u7684 \u8003 \u91cf \uff0c \u56e0 \u6b64 \u5b83 \u5c6c \u65bc \u9451 \u5225 \u5f0f \u8a13 \u7df4 \u7684 \u4e00 \u7a2e \uff0c \u4ee5 \u6700 \u5927 \u4ea4 \u4e92 \u8cc7 \u8a0a \u6cd5 \u5247 \u5f97 \u5230 \u7684 \u6a21 \u578b \u53c3 \u6578 \u53ef \u4f7f \u5f97 \u89c0 \u5bdf \u8cc7 \u6599 X \u8207 \u985e \u5225 m C \u6709 \u8f03 \u9ad8 \u7684 \u76f8 \u4f9d \u6027 \uff0c \u5373 ) , ( X m C I \u039b \u8f03 \u9ad8 \u3002 \u8207 \u6700 \u5c0f \u5206 \u985e \u932f \u8aa4 \u76f8 \u540c \uff0c \u6700 \u5927 \u4ea4 \u4e92 \u8cc7 \u8a0a \u4e5f \u5fc5 \u9808 \u4ee5 \u5ee3 \u7fa9 \u6a5f \u7387 \u905e \u6e1b \u6f14 \u7b97 \u6cd5 \u5be6 \u73fe \uff0c \u5373 ) , ( 1 X m n n C I \u039b + \u2207 \u2212 \u039b = \u039b \u03b5 (11) \u5728 \u9019 \u88e1 \u4ee5 \u8f49 \u79fb \u6a5f \u7387 \u3001 \u5e73 \u5747 \u503c \u5411 \u91cf \u3001 \u5171 \u8b8a \u7570 \u77e9 \u9663 \u4f5c \u8aaa \u660e \uff0c \u800c \u504f \u5fae \u7684 \u5c0d \u8c61 \u4e3b \u8981 \u662f \u6700 \u5927 \u4ea4 \u4e92 \u8cc7 \u8a0a \u4e2d \u7684 \u76f8 \u4f3c \u5ea6 \u51fd \u5f0f ) , | ( ) , | ( 1 ) , | ( log \u039b \u039b \u2202 \u2202 \u039b = \u039b \u039b \u2202 \u2202 m m m C P C P C P X X X (12) \u76f8 \u4f3c \u5ea6 \u51fd \u5f0f \u53ef \u7531 forward-backward probability \u8868 \u793a \u2211\u2211 \u2211 \u2211\u2211 = = = = = \u23ad \u23ac \u23ab \u23a9 \u23a8 \u23a7 = = \u039b T t N j j t j N i ij i T t N j j j m t b a t t t C P 1 1 1 1 1 ) ( ) ( ) ( ) ( ) ( ) , | ( \u03b2 \u03b1 \u03b2 \u03b1 x X (13) \u9032 \u4e00 \u6b65 \u4e4b \u53c3 \u6578 \u4f30 \u6e2c \u904e \u7a0b \u8207 \u7d50 \u679c \uff0c \u8acb \u8a73 \u898b [20]\u3002 2.3 \u4e00 \u822c \u5316 \u6700 \u5c0f \u932f \u8aa4 \u7387 (GMER) \u4e00 \u822c \u5316 \u6700 \u5c0f \u932f \u8aa4 \u7387 \u662f \u7531 Qi Li \u5728 2002 \u5e74 \u6240 \u63d0 \u51fa \uff0c \u4ee5 \u4e0b \u7c21 \u4ecb \u4e00 \u822c \u5316 \u6700 \u5c0f \u932f \u8aa4 \u7387 \u7684 \u7cbe \u795e \u548c \u4f5c \u6cd5 \u3002 \u5728 \u4e00 \u500b \u5177 \u6709 M \u500b \u985e \u5225 \u7684 \u5206 \u985e \u554f \u984c \u88e1 \u9762 \uff0c \u4ee4 \u89c0 \u5bdf \u8cc7 \u6599 X \u5c6c \u65bc \u985e \u5225 m C \uff0c i \u03b1 \u8868 \u793a \u5c07 X \u5206 \u985e \u5230 \u985e \u5225 i C \u7684 \u52d5 \u4f5c \uff0c \u5247 \u53ef \u5b9a \u7fa9 \u4e00 loss function \u70ba \u23a9 \u23a8 \u23a7 = \u2260 = = M m i m i m i C l m i , , 1 , 1 0 ) | ( K \u03b1 (14) \u5c07 \u5206 \u985e \u932f \u8aa4 \u6307 \u5b9a \u4e00 \u500b \u55ae \u4f4d \u7684 loss\uff0c\u82e5\u5206\u985e\u6b63\u78ba\u5247\u4e0d\u6307\u5b9a loss\uff0c\u4ee3\u8868\u5206\u985e\u932f\u8aa4\u7684\u98a8\u96aa(risk)\uff0c\u4e14\u5b9a\u7fa9\u5c0d\u89c0\u5bdf\u8cc7\u6599 X \u63a1 \u53d6 \u52d5 \u4f5c i \u03b1 \u7684 \u5206 \u985e \u932f \u8aa4 \u6a5f \u7387 \u70ba \u2211 = \u2212 = = M j m j j i i C P C P C l R 1 ) | ( 1 ) | ( ) | ( ) | ( X X X \u03b1 \u03b1 (15) ) | ( X m C P \u4ee3 \u8868 X \u5c6c \u65bc \u985e \u5225 m C \u7684 \u4e8b \u5f8c \u6a5f \u7387 \uff0c \u8c9d \u6c0f \u6cd5 \u5247 \u544a \u8a34 \u6211 \u5011 \uff0c \u4ee4 ) | ( X m C P \u6700 \u5927 \u53ef \u964d \u4f4e \u5206 \u985e \u932f \u8aa4 \u7684 \u6a5f \u7387 \uff0c \u7a31 \u4f5c \u6700 \u5c0f \u932f \u8aa4 \u7387 (minimum error rate, MER) \uff0c ) | ( X i C P \u4e00 \u822c \u4ee5 \u4e00 \u7d44 \u5b9a \u7fa9 \u597d \u7684 \u6a21 \u578b \u53c3 \u6578 i \u03bb \u4f86 \u8a08 \u7b97 \uff0c \u5373 ) | ( ) | ( X X i i C P C P \u03bb = \uff0c \u7531 \u65bc \u6a21 \u578b \u53c3 \u6578 \u8207 \u985e \u5225 \u6709 \u4e00 \u5c0d \u4e00 \u7684 \u95dc \u4fc2 \uff0c \u56e0 \u6b64 \u7c21 \u5316 \u8868 \u793a \u70ba ) | ( ) | ( X X i i P C P \u03bb = \u3002 \u5728 \u8a13 \u7df4 \u65b9 \u9762 \uff0c \u9996 \u5148 \u5b9a \u7fa9 \u805a \u96c6 \u4e8b \u5f8c \u6a5f \u7387 (", |
| "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": "M J 1 1 , , ) ( ) | ( 1 X X \u03bb (16) n m, X \u4ee3 \u8868 \u6a21 \u578b m \u7684 \u7b2c n \u500b \u8a13 \u7df4 \u8cc7 \u6599 \uff0c \u9577 \u5ea6 \u70ba n T \uff0c \u5373 { } n T t t n m n m 1 , , , = = x X \uff0c m P \u70ba \u985e \u5225 m \u7684 \u4e8b \u524d \u6a5f \u7387 \uff0c \u5047 \u8a2d \u8a13 \u7df4 \u8cc7 \u6599 \u5206 \u4f48 \u70ba independent, identically distributed (i.i.d) \uff0c \u56e0 \u6b64 n m, X \u8207 m \u03bb \u7684 \u76f8 \u4f3c \u5ea6 \u53ef \u8868 \u793a \u70ba \u220f = = n T t m t n m m n m x P X P 1 , , , ) | ( ) | ( \u03bb \u03bb \u3002 \u70ba \u4e86 \u5177 \u6709 \u9451 \u5225 \u5f0f \u8a13 \u7df4 \u7684 \u5f62 \u5f0f \uff0c \u5c07 (16)\u5f0f\u6539\u5beb\u70ba \u2211\u2211 \u2211\u2211 = = = = = = M m N n n m M m N n n m m m l M d l M J 1 1 , 1 1 , 1 ) ( 1 (17) \u5176 \u4e2d l \u70ba (6)\u5f0f sigmoid function\u3002 \u2211 \u2260 \u2212 = m j j j n m m m n m n m P P P P d ) | ( log ) | ( log , , , \u03bb \u03bb X X (18) \u70ba \u4e86 \u8b93 \u6b63 \u78ba \u985e \u5225 \u8207 \u7af6 \u722d \u985e \u5225 \u4f54 \u6709 \u4e0d \u540c \u7684 \u767e \u5206 \u6bd4 \uff0c \u5728 (18)\u5f0f\u88e1\u7b2c\u4e8c\u9805\u4e58\u4e0a L \uff0c 1 0 \u2264 < L \uff0c \u7576 1 = L \u6642 \uff0c \u4ee3 \u8868 \u6b63 \u78ba \u985e \u5225 \u8207 \u7af6 \u722d \u985e \u5225 \u5177 \u540c \u6a23 \u91cd \u8981 \u6027 \uff0c \u540c \u6642 \u4ee4 (6)\u5f0f sigmoid function \u5167 1 = \u03b3 \uff0c 0 = \u03b8 \u6642 \uff0c J J = \uff0c ) | ( , m n m P \u03bb X \u70ba GMM \u51fd \u5f0f \uff0c \u70ba \u4e86 \u4ee4 J \u70ba \u6700 \u5927 \uff0c \u56e0 \u6b64 \u5c0d J \u53d6 gradient \u4e26 \u4ee4 \u70ba \u96f6 \u53ef \u5f97 \u5230 0 ) | ( log ) ( ) | ( log ) ( 1 1 , , , , , , 1 1 , , , , , , = \u2207 \u2126 \u2212 \u2207 \u2126 = \u2207 \u2211\u2211\u2211 \u2211\u2211 \u2260 = = = = m j N n T t i m t n j t n j i j N n T t i m t n m t n m i m j n mi m n mi mi P L P J \u03bb \u03bb \u03b8 \u03b8 \u03b8 x x x x (19) \u5176 \u4e2d ) | ( ) | ( ) 1 ( ) ( , , , , , , , , , , , m t n m i m t n m i m n m n m t n m i m P P c l l \u03bb \u03bb x x x \u2212 = \u2126 (20) \u2211 \u2260 \u2212 = \u2126 j k k k t n j m i m t n j i m n j n j t n j i j P P P P c l l ) | ( ) | ( ) 1 ( ) ( , , , , , , , , , , , \u03bb \u03bb x x x (21) \u70ba \u4e86 \u5f97 \u5230 \u6a21 \u578b \u53c3 \u6578 \u7684 \u5c01 \u9589 \u89e3 (close-form solution)\uff0c\u9019\u88e1\u5047\u8a2d(20)\u8207(21)\u5f0f\u8207\u6a21\u578b\u53c3\u6578\u7368\u7acb\uff0c\u82e5\u6b32\u6c42\u5e73\u5747\u503c\u5411\u91cf\uff0c\u5c07(19) \u5f0f ) | ( log , , , i m t n m P \u03bb x \u5c0d \u5e73 \u5747 \u503c \u5411 \u91cf \u53d6 \u504f \u5fae \u5206 \u5f8c \u53ef \u5f97 ) ( ) | ( log , , , 1 , , , , , i m t n m i m i m t n m \u00b5 P i m \u00b5 \u03bb \u2212 \u03a3 = \u2207 \u2212 x x (22) \u5c07 \u4e0a \u5f0f \u4ee3 \u5165 (19)\u5f0f\u79fb\u9805\u5f8c\u53ef\u5f97\u5e73\u5747\u503c\u5411\u91cf\u7684\u89e3\u70ba \u2211 \u2211 \u2211 \u2211 \u2211 \u2211 \u2211 \u2211 \u2211 \u2211 = = \u2260 = = \u2260 = = = = \u2126 \u2212 \u2126 \u2126 \u2212 \u2126 = m n j n j n m n N n T t m j N n T t t n j i j t n m i m m j N n T t t n j t n j i j N n T t t n m t n m i m i m L L 1 1 1 1 , , , , , , 1 1 , , , , , 1 1 , , , ,", |
| "eq_num": ", , ) ( )" |
| } |
| ], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "( ) ( ) ( x x x x x x \u00b5 (23) 2.4 \u7dda \u6027 \u8ff4 \u6b78 \u8a9e \u8005 \u8abf \u9069 \u6839 \u64da \u8a9e \u97f3 \u6a21 \u578b \u8207 \u8a9e \u8005 \u9593 \u4e4b \u76f8 \u95dc \u6027 \u53ef \u5206 \u70ba \u8a9e \u8005 \u7368 \u7acb (speaker-independent, SI) \u8a9e \u97f3 \u6a21 \u578b \u53ca \u8a9e \u8005 \u76f8 \u4f9d (speaker-dependent, SD)\u8a9e\u97f3\u6a21\u578b\u3002\u4f7f\u7528\u8a9e\u8005\u76f8\u4f9d\u4e4b\u8a9e\u97f3\u6a21\u578b\uff0c\u5728\u8fa8\u8b58\u6642\uff0c\u9808\u5148\u884c\u6307\u5b9a\u6216\u5075\u6e2c\u8981\u4f7f\u7528\u7684\u8a9e\u8005\u6a21\u578b\u7d44\u5225\uff0c \u800c \u8a9e \u8005 \u7368 \u7acb \u5247 \u4e0d \u9808 \uff0c \u4ee5 \u6b64 \u5dee \u5225 \u770b \u4f86 \uff0c \u8a9e \u8005 \u76f8 \u4f9d \u4e4b \u8a9e \u97f3 \u8fa8 \u8b58 \u7cfb \u7d71 \uff0c \u4f7f \u7528 \u4e0a \u8f03 \u4e0d \u4fbf \uff0c \u4e14 \u9700 \u5132 \u5b58 \u591a \u7d44 \u8a9e \u97f3 \u6a21 \u578b \u3002 \u76f8 \u5c0d \u4f86 \u8aaa \uff0c \u4f7f \u7528 \u8a9e \u8005 \u7368 \u7acb \u8a9e \u97f3 \u6a21 \u578b \u6642 \u6240 \u9700 \u8981 \u7684 \u8a9e \u97f3 \u6a21 \u578b \u6578 \u91cf \u6703 \u8f03 \u5c11 \u4e14 \u6a21 \u578b \u7279 \u6027 \u8207 \u6bcf \u4e00 \u4f4d \u6e2c \u8a66 \u8a9e \u8005 \u5747 \u4e0d \u751a \u543b \u5408 \u3002 \u6240 \u4ee5 \uff0c \u8fa8 \u8b58 \u7387 \u6703 \u8f03 \u5dee \u3002 \u4e00 \u822c \u800c \u8a00 \uff0c \u4f7f \u7528 \u8a9e \u8005 \u76f8 \u4f9d \u8a9e \u97f3 \u6a21 \u578b \u7684 \u8fa8 \u8b58 \u7cfb \u7d71 \u6548 \u80fd \u6703 \u6bd4 \u8a9e \u8005 \u7368 \u7acb \u4e4b \u8fa8 \u8b58 \u7cfb \u7d71 \u6548 \u80fd \u9ad8 \u4e8c \u81f3 \u4e09 \u500d [7]\u3002 \u70ba \u4e86 \u4fdd \u7559 \u5169 \u8005 \u512a \u9ede \uff0c \u4e00 \u822c \u7686 \u8a13 \u7df4 \u51fa \u4e00 \u7d44 \u8a9e \u8005 \u7368 \u7acb \u7684 \u8a9e \u97f3 \u6a21 \u578b \uff0c \u53d6 \u5176 \u6a21 \u578b \u7e3d \u6578 \u91cf \u8f03 \u5c11 \u7684 \u512a \u9ede \uff0c \u800c \u4ee5 \u6b64 \u6a21 \u578b \u70ba \u57fa \u790e \uff0c \u518d \u5229 \u7528 \u4e00 \u4e9b \u7531 \u6e2c \u8a66 \u8a9e \u8005 \u6240 \u9304 \u5f97 \u4e4b \u8abf \u9069 \u8a9e \u6599 \uff0c \u5148 \u8abf \u9069 \u51fa \u8207 \u8a72 \u8a9e \u8005 \u8a9e \u97f3 \u7279 \u6027 \u8f03 \u76f8 \u7b26 \u7684 \u8a9e \u97f3 \u6a21 \u578b \uff0c \u5373 \u6240 \u8b02 \u7684 \u8a9e \u8005 \u76f8 \u4f9d \u8a9e \u97f3 \u6a21 \u578b \uff0c \u53ef \u6709 \u6548 \u63d0 \u5347 \u8a9e \u97f3 \u8fa8 \u8b58 \u7387 \u3002 \u4e0d \u904e \u7528 \u65bc \u8abf \u6574 \u7684 \u8a9e \u6599 \u4e00 \u822c \u4e26 \u4e0d \u591a \uff0c \u5bb9 \u6613 \u9020 \u6210 \u8abf \u9069 \u8a9e \u6599 \u7a00 \u758f \u7684 \u554f \u984c \uff0c \u70ba \u4e86 \u89e3 \u6c7a \u6a23 \u672c \u6578 \u4e0d \u8db3 \u7684 \u554f \u984c \uff0c \u505a \u6cd5 \u662f \u5c07 \u8a9e \u97f3 \u6a21 \u578b \u5206 \u7fa4 \uff0c \u70ba \u6bcf \u4e00 \u7fa4 \u7684 \u8a9e \u97f3 \u6a21 \u578b \u627e \u51fa \u4e00 \u500b \u53c3 \u6578 \u8f49 \u63db \u77e9 \u9663 \uff0c \u7fa4 \u96c6 \u5167 \u7684 \u6a21 \u578b \u8abf \u6574 \u53ea \u8981 \u4f9d \u7167 \u6b64 \u8f49 \u63db \u77e9 \u9663 \u5373 \u53ef \u5f97 \u5230 \u66f4 \u65b0 \u5f8c \u53c3 \u6578 \u3002 \u70ba \u4e86 \u5f97 \u5230 \u66f4 \u65b0 \u5f8c \u7684 \u8f49 \u63db \u77e9 \u9663 \uff0c \u53ef \u4ee5 \u5229 \u7528 \u4e0d \u540c \u7684 \u6cd5 \u5247 \uff0c \u8f03 \u5e38 \u898b \u7684 \u6709 \u6700 \u5927 \u76f8 \u4f3c \u5ea6 \u7dda \u6027 \u8ff4 \u6b78 \u6cd5 \u5247 \uff0c \u6700 \u5927 \u4e8b \u5f8c \u6a5f \u7387 \u7dda \u6027 \u8ff4 \u6b78 \u6cd5 \u5247 \uff0c \u6700 \u5c0f \u5206 \u985e \u932f \u8aa4 \u7dda \u6027 \u8ff4 \u6b78 \u6cd5 \u5247 \u3002 2.5 \u6700 \u5927 \u76f8 \u4f3c \u5ea6 \u7dda \u6027 \u8ff4 \u6b78 (MLLR) \u6700 \u5927 \u76f8 \u4f3c \u5ea6 \u7dda \u6027 \u8ff4 \u6b78 \u7684 \u76ee \u6a19 \u5c31 \u662f \uff0c \u5c0d \u4e00 \u7fa4 \u96c6 s\uff0c\u8a08\u7b97\u4e00\u8f49\u63db\u77e9\u9663 s W \uff0c \u4f7f \u5f97 \u7fa4 \u96c6 \u5167 \u6240 \u6709 \u8abf \u9069 \u8cc7 \u6599 \u7684 \u76f8 \u4f3c \u5ea6 \u6700 \u5927 \uff0c \u6700 \u5927 \u76f8 \u4f3c \u5ea6 \u7dda \u6027 \u8ff4 \u6b78 \u8abf \u9069 \u6f14 \u7b97 \u6cd5 \u7684 \u597d \u8655 \u5728 \u65bc \uff0c \u8abf \u9069 \u8a9e \u6599 \u4e0d \u9700 \u8981 \u5b8c \u5168 \u6db5 \u84cb \u6240 \u6709 \u6a21 \u578b \uff0c \u5373 \u4f7f \u6c92 \u6709 \u8abf \u9069 \u8cc7 \u6599 \u7684 \u6a21 \u578b \uff0c \u4e5f \u53ef \u4ee5 \u7d93 \u7531 \u540c \u985e \u5225 \u7684 \u8f49 \u63db \u77e9 \u9663 \u9032 \u884c \u8abf \u9069 \u3002 \u4ee5 \u8abf \u6574 \u5e73 \u5747 \u503c \u5411 \u91cf \u70ba \u4f8b \uff0c \u5728 \u8a08 \u7b97 \u8f49 \u63db \u77e9 \u9663 \u4e4b \u524d \uff0c \u5c07 \u5e73 \u5747 \u503c \u5411 \u91cf \u5ef6 \u5c55 \u70ba [ ] T D s \u00b5 \u00b5 \u00b5 \u03be , , , , 1 2 1 K = (24) \u5176 \u4e2d \uff0c D \u70ba \u5411 \u91cf \u7dad \u5ea6 \uff0c \u5247 \u66f4 \u65b0 \u5f8c \u7684 \u5e73 \u5747 \u503c \u5411 \u91cf \u70ba s s r s \u03be \u00b5 ) ( W = (25) \u5176 \u4e2d \uff0c ) (s r \u4ee3 \u8868 \u72c0 \u614b s \u6240 \u5c6c \u8ff4 \u6b78 \u985e \u5225 \uff0c ) (s r W \u4ee3 \u8868 \u8ff4 \u6b78 \u985e \u5225 (regression class) ) (s r \u7684 \u8f49 \u63db \u77e9 \u9663 \uff0c \u7dad \u5ea6 \u70ba ) 1 ( + \u00d7 D D \uff0c \u5247 \u900f \u904e EM \u6f14 \u7b97 \u6cd5 \uff0c \u6700 \u5f8c \u53ef \u4ee5 \u5f97 \u5230 \u6bcf \u4e00 \u500b \u8ff4 \u6b78 \u985e \u5225 \u7684 \u8f49 \u63db \u77e9 \u9663 \u4e4b \u6bcf \u4e00 \u5217 \u8a08 \u7b97 \u65b9 \u5f0f \u5982 \u4e0b T i i T i z G w 1 ) ( \u2212 = (26) T i w \u548c T i z \u5206 \u5225 \u4ee3 \u8868 W \u548c Z \u7684 \u5217 \u5411 \u91cf [14]\u3002 2.6 \u6700 \u5927 \u4e8b \u5f8c \u6a5f \u7387 \u7dda \u6027 \u8ff4 \u6b78 (MAPLR) \u7531 \u65bc \u4ee5 \u6700 \u5927 \u76f8 \u4f3c \u5ea6 \u70ba \u4e3b \u4e4b \u7dda \u6027 \u8f49 \u63db \u77e9 \u9663 \u5728 \u8a08 \u7b97 \u4e0a \u5341 \u5206 \u7c21 \u6613 \uff0c \u6240 \u4ee5 \u5176 \u61c9 \u7528 \u5341 \u5206 \u666e \u904d \uff0c \u7136 \u800c \uff0c \u82e5 \u8abf \u9069 \u8a9e \u6599 \u904e \u5c11 \uff0c \u6216 \u8a9e \u6599 \u7279 \u6027 \u4e0d \u5177 \u4ee3 \u8868 \u6027 \u6642 \uff0c \u5247 \u53ef \u80fd \u5c0e \u81f4 \u5f97 \u5230 \u7684 \u8f49 \u63db \u77e9 \u9663 \u4ecd \u820a \u7121 \u6cd5 \u7b26 \u5408 \u6e2c \u8a66 \u8a9e \u8005 \u7684 \u8a9e \u97f3 \u7279 \u6027 \uff0c \u65bc \u662f \uff0c \u4fbf \u8003 \u616e \u5230 \u5f15 \u5165 \u8f49 \u63db \u77e9 \u9663 \u7684 \u4e8b \u524d \u5206 \u4f48 \u8cc7 \u8a0a \u3002 \u77e9 \u9663 \u53c3 \u6578 \u7684 \u4e8b \u524d \u5206 \u4f48 \u53ef \u4ee5 \u5728 \u4f30 \u6e2c \u8f49 \u63db \u77e9 \u9663 \u6642 \u9650 \u5236 \u53c3 \u6578 \u53ef \u80fd \u7684 \u8abf \u9069 \u91cf \uff0c \u4f7f \u5f97 \u53c3 \u6578 \u7684 \u4f30 \u6e2c \u66f4 \u5177 \u5f37 \u5065 \u6027 \uff0c \u7531 \u6587 \u737b \u5be6 \u9a57 \u53ef \u770b \u51fa \uff0c \u6700 \u5927 \u4e8b \u5f8c \u6a5f \u7387 \u7dda \u6027 \u8ff4 \u6b78 \u53ef \u9054 \u5230 \u6bd4 \u6700 \u5927 \u76f8 \u4f3c \u5ea6 \u7dda \u6027 \u8ff4 \u6b78 \u66f4 \u597d \u7684 \u8fa8 \u8b58 \u7387 [21]\u3002 2.7 \u6700 \u5c0f \u5206 \u985e \u932f \u8aa4 \u7dda \u6027 \u8ff4 \u6b78 (MCELR) \u6700 \u5c0f \u5206 \u985e \u932f \u8aa4 \u7684 \u9451 \u5225 \u5f0f \u8a13 \u7df4 \u65b9 \u5f0f \u5728 \u5f88 \u591a \u61c9 \u7528 \u90fd \u80fd \u986f \u793a \u51fa \u4e0d \u932f \u7684 \u6548 \u80fd \uff0c \u4e0d \u904e \u6700 \u5c0f \u5206 \u985e \u932f \u8aa4 \u4e00 \u822c \u4ee5 \u5ee3 \u7fa9 \u6a5f \u7387 \u905e \u6e1b \u6f14 \u7b97 \u6cd5 \u5be6 \u73fe \uff0c \u4e26 \u6c92 \u6709 \u5728 \u7406 \u8ad6 \u4e0a \u8b49 \u660e \u5b83 \u80fd \u6536 \u6582 \u5230 \u66f4 \u597d \u7684 \u6a21 \u578b \uff0c \u7576 \u8a13 \u7df4 \u8cc7 \u6599 \u8b8a \u5c11 \u6642 \uff0c \u932f \u8aa4 \u7684 \u6536 \u6582 \u505c \u6b62 \u9ede \u66f4 \u5bb9 \u6613 \u767c \u751f \uff0c \u56e0 \u6b64 \u5c07 MCE \u61c9 \u7528 \u5728 \u6a21 \u578b \u8abf \u9069 \u6642 \uff0c \u4f7f \u7528 \u7dda \u6027 \u8ff4 \u6b78 \u6709 \u5176 \u5fc5 \u8981 \u3002 Chengalvarayan \u5728 1998 \u5e74 \u63d0 \u51fa \u6700 \u5c0f \u5206 \u985e \u932f \u8aa4 \u7dda \u6027 \u8ff4 \u6b78 [4]\uff0c \u4f7f \u7528 \u5168 \u57df \u6027 \u7684 \u8f49 \u63db \u77e9 \u9663 \u4e26 \u4ee5 \u5ee3 \u7fa9 \u6a5f \u7387 \u905e \u6e1b \u6f14 \u7b97 \u6cd5 \u4f30 \u6e2c \u77e9 \u9663 \u53c3 \u6578 \uff0c \u5be6 \u9a57 \u7d50 \u679c \u986f \u793a \u51fa \u5176 \u8abf \u9069 \u6548 \u679c \u6bd4 \u6700 \u5927 \u76f8 \u4f3c \u5ea6 \u7dda \u6027 \u8ff4 \u6b78 \u6f14 \u7b97 \u6cd5 \u597d \u3002 \u800c \u5728 [10]\u4e2d\uff0c\u66f4\u9032\u4e00\u6b65\u4f7f\u7528\u591a\u7d44\u8ff4\u6b78\u985e\u5225\u7684\u8f49\u63db\u77e9\u9663\u9032\u884c\u8abf\u9069\uff0c\u5728\u540c\u6a23\u4f7f\u7528\u5ee3\u7fa9\u6a5f\u7387\u905e\u6e1b\u6f14\u7b97\u6cd5\u4e0b\uff0c\u53ef \u4ee5 \u6709 \u66f4 \u597d \u7684 \u8abf \u9069 \u6548 \u80fd \u6539 \u9032 \u3002 \u53e6 \u5916 \uff0c \u5728 [9]\u4e2d\uff0c\u4f5c\u8005\u4e0d\u5229\u7528\u5ee3\u7fa9\u6a5f\u7387\u905e\u6e1b\u6f14\u7b97\u6cd5\u5be6\u73fe\u6700\u5c0f\u5206\u985e\u932f\u8aa4\u7dda\u6027\u8ff4\u6b78\u8abf\u9069\u6f14\u7b97\u6cd5\uff0c \u800c \u4ee5 \u4e00 \u822c \u5316 \u8abf \u9069 \u4f5c \u6cd5 \u8a08 \u7b97 \u8f49 \u63db \u77e9 \u9663 \uff0c \u5373 \u8f49 \u63db \u77e9 \u9663 \u4ee5 \u7fa4 \u96c6 \u70ba \u55ae \u4f4d \uff0c \u5c07 \u6700 \u5c0f \u5206 \u985e \u932f \u8aa4 \u7684 \u76ee \u6a19 \u51fd \u5f0f \u6539 \u5beb \u5f8c \uff0c \u53ef \u4ee5 \u900f \u904e EM \u6f14 \u7b97 \u6cd5 \u4ee5 \u5c01 \u9589 \u89e3 \u7684 \u65b9 \u5f0f \u8a08 \u7b97 \u8f49 \u63db \u77e9 \u9663 \u3002 3. \u805a\u96c6\u4e8b\u5f8c\u6a5f\u7387\u7dda\u6027\u8ff4\u6b78\u9451\u5225\u5f0f\u8abf\u9069\u6cd5 \u5728 \u6700 \u5c0f \u5206 \u985e \u932f \u8aa4 \u4f30 \u6e2c \u6cd5 \u5247 \u4e2d \uff0c \u4e26 \u4e0d \u8003 \u616e \u985e \u5225 \u7684 \u4e8b \u524d \u8cc7 \u8a0a \uff0c \u4e14 \u4f7f \u7528 \u5ee3 \u7fa9 \u6a5f \u7387 \u905e \u6e1b \u6f14 \u7b97 \u6cd5 \u5be6 \u73fe \uff0c \u5728 \u8abf \u9069 \u8cc7 \u6599 \u5c11 \u6642 \uff0c \u66f4 \u5bb9 \u6613 \u767c \u751f \u932f \u8aa4 \u8a13 \u7df4 \u7684 \u554f \u984c \uff0c \u56e0 \u6b64 \uff0c Beyerlin \u5c07 \u6240 \u6709 \u6a21 \u578b (\u8a9e\u97f3\u6a21\u578b\u3001\u8a9e\u8a00\u6a21\u578b)\u7d44\u6210\u4e00\u500b\u4e8b\u5f8c\u6a5f\u7387\u7684\u7dda\u6027\u7d44\u5408\uff0c\u5229\u7528 \u9451 \u5225 \u5f0f \u8a13 \u7df4 \u4f30 \u6e2c \u51fa \u7dda \u6027 \u7d44 \u5408 \u7684 \u4fc2 \u6578 [2]", |
| "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": "\u8003 \u616e \u5230 \u6700 \u5927 \u4e8b \u5f8c \u6a5f \u7387 \u5728 \u5c11 \u91cf \u8a13 \u7df4 \u8a9e \u6599 \u4e0b \u53ef \u4ee5 \u5f97 \u5230 \u6bd4 \u6700 \u5927 \u76f8 \u4f3c \u5ea6 \u8f03 \u6b63 \u78ba \u7684 \u6a21 \u578b \u53c3 \u6578 \uff0c \u7531 \u524d \u8ff0 \u7684 \u4e00 \u822c \u5316 \u6700 \u5c0f \u932f \u8aa4 \u7387 \u4ecb \u7d39 \u4e2d \u53ef \u4ee5 \u770b \u51fa \uff0c \u5b83 \u5c07 \u4e8b \u5f8c \u6a5f \u7387 \u4e2d \u539f \u672c \u8207 \u6a21 \u578b \u53c3 \u6578 \u7121 \u95dc \u7684 ) ( ,n m P x \u8868 \u793a \u6210 \u8207 \u6a21 \u578b \u76f8 \u95dc \uff0c \u5373 \u5177 \u9451 \u5225 \u5f0f \u8a13 \u7df4 \u7684 \u5f62 \u5f0f \uff0c \u5c07 \u539f \u672c \u6700 \u5c0f \u5206 \u985e \u932f \u8aa4 \u4e2d \u9451 \u5225 \u5f0f \u51fd \u5f0f \u70ba \u76f8 \u4f3c \u5ea6 \u51fd \u5f0f \u6539 \u70ba \u4e8b \u5f8c \u6a5f \u7387 \u51fd \u5f0f \uff0c \u53ef \u4ee5 \u7d50 \u5408 \u9019 \u5169 \u7a2e \u6a21 \u578b \u4f30 \u6e2c \u65b9 \u5f0f \u7684 \u512a \u9ede \uff0c \u4e26 \u5229 \u7528 \u5c01 \u9589 \u89e3 \u7684 \u89e3 \u6cd5 \u53ef \u4ee5 \u5feb \u901f \u4f30 \u6e2c \u51fa \u6a21 \u578b \u53c3 \u6578 \uff0c \u6539 \u5584 \u4ee5 \u5f80 \u4ee5 \u5ee3 \u7fa9 \u6a5f \u7387 \u905e \u6e1b \u6cd5 \u5247 \u5be6 \u4f5c \u6642 \u6536 \u6582 \u592a \u6162 \u7684 \u7f3a \u9ede \u3002 \u7531 \u65bc \u8a9e \u97f3 \u6a21 \u578b \u8abf \u9069 \u6642 \u8cc7 \u6599 \u91cf \u901a \u5e38 \u8f03 \u5c11 \uff0c \u56e0 \u6b64 \u5c07 \u4e00 \u822c \u5316 \u6700 \u5c0f \u932f \u8aa4 \u7387 \u7684 \u65b9 \u5f0f \u5c0e \u5165 \u5c07 \u6709 \u52a9 \u65bc \u53c3 \u6578 \u7684 \u4f30 \u6e2c \uff0c \u6211 \u5011 \u5c07 \u6b64 \u8abf \u9069 \u7684 \u65b9 \u5f0f \u7a31 \u70ba \u805a \u96c6 \u4e8b \u5f8c \u6a5f \u7387 \u7dda \u6027 \u8ff4 \u6b78 (AAPLR)\u8abf\u9069\u6f14\u7b97\u6cd5\u3002\u70ba\u4e86\u4ee5 AAPLR \u7684 \u65b9 \u5f0f \u8a08 \u7b97 \u8f49 \u63db \u77e9 \u9663 \uff0c \u4e14 \u52a0 \u5165 \u8f49 \u63db \u77e9 \u9663 \u7684 \u4e8b \u524d \u8cc7 \u8a0a \u53ef \u4ee5 \u8b93 \u5176 \u4f30 \u6e2c \u8f03 \u5177 \u5f37 \u5065 \u6027 \uff0c \u56e0 \u6b64 \u5c07", |
| "eq_num": "(" |
| } |
| ], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "EQUATION", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [ |
| { |
| "start": 0, |
| "end": 8, |
| "text": "EQUATION", |
| "ref_id": "EQREF", |
| "raw_str": "\u5728 \u7e7c \u7e8c \u63a8 \u5c0e \u805a \u96c6 \u4e8b \u5f8c \u6a5f \u7387 \u7dda \u6027 \u8ff4 \u6b78 \u6f14 \u7b97 \u6cd5 \u524d \uff0c \u6211 \u5011 \u5c07 \u900f \u904e EM \u6f14 \u7b97 \u6cd5 \uff0c \u767c \u6398 \u6700 \u5927 \u4e8b \u5f8c \u6a5f \u7387 \u7dda \u6027 \u8ff4 \u6b78 \u8207 \u4f7f \u7528 \u6700 \u5c0f \u5206 \u985e \u932f \u8aa4 \u6e96 \u5247 \u4e4b \u53c3 \u6578 \u4f30 \u6e2c \u6f14 \u7b97 \u6cd5 \u4e4b \u9593 \u7684 \u5dee \u7570 \u3002 \u7d66 \u5b9a \u4e00 \u8a9e \u97f3 \u89c0 \u5bdf \u6a23 \u672c \u5e8f \u5217 { } T x x X , , 1 K = \uff0c \u5176 \u9577 \u5ea6 \u70ba T \uff0c \u4e14 \u5b58 \u5728 \u7dda \u6027 \u8f49 \u63db \u77e9 \u9663 \u96c6 \u5408 { } R w w W , , 1 K = \uff0c \u5176 \u4e2d \u5171 \u6709 R \u7d44 \u985e \u5225 \u3002 \u5247 \u5728 \u7d66 \u5b9a \u89c0 \u5bdf \u6a23 \u672c \u5e8f \u5217 X \u6642 \uff0c \u7dda \u6027 \u8f49 \u63db \u77e9 \u9663 \u7684 \u4e8b \u5f8c \u6a5f \u7387 \u53ef \u8868 \u793a \u5982 \u4e0b ) ; | ( \u039b X W g (28) \u5176 \u4e2d \uff0c \u039b \u8868 \u793a \u7528 \u65bc \u76f8 \u4f3c \u5ea6 \u8a08 \u7b97 \u4e4b \u8a9e \u97f3 \u6a21 \u578b \u96c6 \u5408 \u3002 \u800c \u4e0a \u8ff0 \u4e4b \u4e8b \u5f8c \u6a5f \u7387 \u53c8 \u53ef \u4ee5 \u900f \u904e \u8c9d \u6c0f \u6cd5 \u5247 \u8f49 \u63db \u5982 \u4e0b \u4e4b \u76f8 \u4f3c \u5ea6 \u8207 \u4e8b \u524d \u6a5f \u7387 \u4e4b \u7d44 \u5408 \u220f = \u039b = \u039b = \u039b T t t t p g p p g p g 1 ) ( ) ( ) ; | ( ) ( ) ( ) ; | ( ) ; | ( x W W x X W W X X W (29) \u6b64 \u8655 \u4e4b ) (W g \u4ee3 \u8868 \u7dda \u6027 \u8f49 \u63db \u77e9 \u9663 W \u4e4b \u4e8b \u524d \u5206 \u4f48 \u6a5f \u7387 \u3002 \u518d \u9032 \u4e00 \u6b65 \u5c07 (29)\u5f0f\u5c0d\u6578\u5316\u53ef\u5f97 \u2211 = \u039b = \u039b T t t t x p W g W x p X W g 1 ) ( ) ( ) ; | ( log ) ; | ( log (30) \u5728 EM \u6f14 \u7b97 \u6cd5 \u4e2d \u4e4b E-step \u5373 \u7528 \u65bc \u8a08 \u7b97 \u4ee5 \u4e0b \u4e4b \u8f14 \u52a9 \u51fd \u5f0f \u2211\u2211 = = \u039b = \u039b = = \u23aa \u23ad \u23aa \u23ac \u23ab \u23aa \u23a9 \u23aa \u23a8 \u23a7 \u039b = M i T t t t t t t p g i q p i q p p g p E R 1 1 ) ( ) ( ) ; | , ( log ) ; , | ( , ) ( ) ( ) ; | , ( log ) | ( x W W x W x W X X W W q X W W (31) \u5176 \u4e2d \uff0c ( ) T q q q , , , 2 1 K = q \u8868 \u793a \u7d66 \u5b9a \u4e4b \u89c0 \u5bdf \u5e8f \u5217 X \u4e4b \u6bcf \u4e00 \u6642 \u9593 \u9ede \u6240 \u5c0d \u61c9 \u4e4b \u72c0 \u614b \u5e8f \u5217 \u3002 \u0174 \u8868 \u793a \u900f \u904e EM \u6f14 \u7b97 \u6cd5 \u6b32 \u4f30 \u6e2c \u4e4b \u65b0 \u8f49 \u63db \u77e9 \u9663 \u53c3 \u6578 \uff0c \u800c W \u5247 \u662f \u73fe \u6709 \u900f \u904e EM \u6f14 \u7b97 \u6cd5 \u5728 \u524d \u4e00 \u6b21 M \u6b65 \u9a5f \u4e2d \u6240 \u4f30 \u6e2c \u51fa \u4e4b \u6700 \u4f73 \u8f49 \u63db \u77e9 \u9663 \u53c3 \u6578 \u3002 \u4ee4 ) ; , | ( ) ( \u039b = = W x x t t t i i q p \u03b3 \u7528 \u4ee5 \u8868 \u793a \u5728 \u7b2c t \u500b \u6642 \u9593 \u9ede \uff0c \u89c0 \u5bdf \u6a23 \u672c t x \u505c \u7559 \u65bc \u7b2c i \u500b \u72c0 \u614b \u4e4b \u6a5f \u7387 \uff0c \u5247 (31)\u5f0f\u4e4b\u8f14\u52a9 \u51fd \u5f0f \u53ef \u7c21 \u55ae \u8868 \u793a \u70ba \u2211\u2211 = = \u039b = = \u23aa \u23ad \u23aa \u23ac \u23ab \u23aa \u23a9 \u23aa \u23a8 \u23a7 \u039b = M i T t t t t t i p g i q p p g p E R 1 1 ) ( ) ( ) ; | , ( log ) ( , ) ( ) ( ) ; | , ( log ) | ( x W W x x W X X W W q X W W \u03b3 . (32) \u5728 \u6b64 \uff0c \u6211 \u5011 \u4f7f \u7528 \u7dad \u7279 \u6bd4 (Viterbi)\u8fd1\u4f3c\u6cd5\u5247\u4f86\u7c21\u5316\u6211\u5011\u7684\u5f0f\u5b50\u3002\u65bc\u662f\uff0c\u6211\u5011\u4f7f\u7528\u6700\u4f73\u7684\u72c0\u614b\u5e8f\u5217\u4f86\u53d6\u4ee3\u539f\u6709\u9700\u8003\u616e\u6240\u6709 \u53ef \u80fd \u6027 \u4e4b \u8868 \u793a \u6cd5 \uff0c \u540c \u6642 \u4e0a \u8ff0 \u4e4b \u72c0 \u614b \u505c \u7559 \u6a5f \u7387 ) ( t i x \u03b3 \u5247 \u7c21 \u5316 \u5982 \u4e0b \u23a9 \u23a8 \u23a7 \u2260 = = i q i q t t t i 1 0 ) (x \u03b3 (33) \u6b64 \u5916 \uff0c \u70ba \u4e86 \u8207 \u4ee5 \u4e0b \u4e4b \u805a \u96c6 \u4e8b \u5f8c \u6a5f \u7387 \u6bd4 \u8f03 \uff0c \u6211 \u5011 \u5c07 \u4f7f \u7528 \u4e0b \u5217 \u5b9a \u7fa9 \u4e4b \u7b26 \u865f \u91cd \u65b0 \u8868 \u793a (32)\u5f0f\u3002\u4f7f\u7528 m \u4f86 \u8868 \u793a \u539f \u6709 \u4e4b \u72c0 \u614b \u6a19 \u793a i \uff0c \u5373 \u5c07 \u72c0 \u614b \u8996 \u70ba \u8a9e \u97f3 \u6a21 \u578b \u985e \u5225 \uff1b \u4f7f \u7528 n m, x \u53d6 \u4ee3 \u539f \u6709 \u4e4b t x \u3002 \u56e0 \u70ba \u539f \u6709 \u4e4b \u89c0 \u5bdf \u6a23 \u672c t x \u5728 \u7d93 \u904e \u7dad \u7279 \u6bd4 \u89e3 \u78bc \u5668 \u5c0d \u61c9 \u51fa \u6700 \u4f73 \u4e4b \u72c0 \u614b \u5f8c \uff0c \u5373 \u53ef \u4ee5 \u660e \u78ba \u77e5 \u9053 \u5169 \u8005 \u9593 \u4e4b \u95dc \u9023 \u3002 \u6240 \u4ee5 \u7528 n m, x \u4f86 \u8868 \u793a \u539f \u6709 \u89c0 \u5bdf \u6a23 \u672c t x \u70ba \u5c0d \u61c9 \u81f3 \u7b2c m \u985e \u6a21 \u578b \u4e4b \u7b2c n \u500b \u89c0 \u5bdf \u6a23 \u672c \u3002 \u5247 (32)\u5f0f\u53ef\u4ee5\u91cd\u65b0\u8868\u793a\u70ba \u2211\u2211 = = \u039b = M m N n n m m r m r n m m p g m p R 1 1 , ) ( ) ( , ) ( ) ( ) ; | , ( log ) | ( x W W x W W . (34) \u5176 \u4e2d \uff0c ) ( m r W \u8868 \u793a \u8a72 \u7dda \u6027 \u8f49 \u63db \u77e9 \u9663 \u662f \u7528 \u65bc \u8f49 \u63db \u7b2c m \u985e \u8a9e \u97f3 \u6a21 \u578b \u53c3 \u6578 \u4e4b \u7528 \u3002 \u4e00 \u822c \u800c \u8a00 \uff0c \u7dda \u6027 \u8f49 \u63db \u77e9 \u9663 \u662f \u6839 \u64da \u6240 \u6709 \u8a9e \u97f3 \u6a21 \u578b \u53c3 \u6578 \u4e2d \u5177 \u76f8 \u4f3c \u7279 \u6027 \u4e4b \u5206 \u7fa4 \u7d50 \u679c \u800c \u5206 \u70ba \u6578 \u500b \u985e \u5225 \uff0c \u5982 \u5206 \u70ba R \u7fa4 \uff0c \u88ab \u5206 \u65bc \u540c \u7fa4 \u4e4b \u8a9e \u97f3 \u6a21 \u578b \u662f \u5171 \u7528 \u540c \u4e00 \u7d44 \u8f49 \u63db \u77e9 \u9663 \u9032 \u884c \u8f49 \u63db \u3002 \u65bc \u662f \u5728 \u7d66 \u5b9a \u8a9e \u97f3 \u6a21 \u578b \u985e \u5225 m \u5f8c \uff0c \u5373 \u53ef \u4ee5 \u900f \u904e \u4e0a \u8ff0 \u4e4b \u95dc \u4fc2 \uff0c \u5f97 \u5230 \u5c0d \u61c9 \u4e4b \u8f49 \u63db \u77e9 \u9663 \u985e \u5225 \u3002 \u53e6 \u5916 \uff0c \u6211 \u5011 \u662f \u4ee5 ) (m r \u8868 \u793a \u7b2c r \u985e \u8f49 \u63db \u77e9 \u9663 \u8207 \u7b2c m \u985e \u8a9e \u97f3 \u6a21 \u578b \u4e4b \u95dc \u4fc2 \u3002 \u5f9e \u53e6 \u4e00 \u65b9 \u9762 \u4f86 \u770b \uff0c \u9075 \u5faa \u4e0a \u8ff0 \u8b8a \u6578 \u3001 \u6a19 \u793a \u4e4b \u5b9a \u7fa9 \uff0c \u5247 \u8f49 \u63db \u77e9 \u9663 W \u4e4b \u805a \u96c6 \u4e8b \u5f8c \u6a5f \u7387 \u5b9a \u7fa9 \u5373 \u70ba (27)\u5f0f\uff0c\u5f9e(27)\u5f0f\u8207(34) \u5f0f \u6bd4 \u8f03 \u53ef \u77e5 \uff0c \u5728 \u4f7f \u7528 EM \u6f14 \u7b97 \u6cd5 \u5c0d \u8a9e \u97f3 \u6a21 \u578b \u6216 \u662f \u6b64 \u8655 \u6240 \u8003 \u616e \u4e4b \u8f49 \u63db \u77e9 \u9663 \u4e4b \u53c3 \u6578 \u9032 \u884c \u4f30 \u6e2c \u6642 \uff0c \u662f \u5c07 \u7b2c (34)\u5f0f\u4e4b ) | ( W W R \u91dd \u5c0d \u6240 \u6b32 \u4f30 \u6e2c \u4e4b \u53c3 \u6578 \u4e88 \u4ee5 \u504f \u5fae \u5206 \u5f8c \uff0c \u800c \u900f \u904e \u5c01 \u9589 \u89e3 \u4f86 \u5f97 \u5230 \u66f4 \u65b0 \u7684 \u53c3 \u6578 \u5167 \u5bb9 \u3002 \u800c \u5728 \u805a \u96c6 \u4e8b \u5f8c \u6a5f \u7387 \u7684 \u5b9a \u7fa9 \u5f0f \u4e2d \uff0c \u5247 \u662f \u5c07 \u5404 \u500b \u985e \u5225 \u4e4b \u4e8b \u5f8c \u6a5f \u7387 \u5168 \u90e8 \u52a0 \u7e3d \u8d77 \u4f86 \uff0c \u65bc \u662f \u5728 \u6587 \u737b \u4e2d \u63a5 \u4e0b \u4f86 \u7684 \u63a8 \u5c0e \u904e \u7a0b \u4e2d \uff0c \u624d \u53ef \u671d \u6240 \u8b02 \u7684 \u6700 \u5c0f \u5206 \u985e \u932f \u8aa4 \u4e4b \u9451 \u5225 \u5f0f \u53c3 \u6578 \u4f30 \u6e2c \u4e4b \u540c \u7406 \u6027 \u9032 \u884c \u63a8 \u5c0e \uff0c \u4e26 \u7d93 \u4e00 \u4e9b \u5047 \u8a2d \u8a2d \u5b9a \u5f8c \uff0c \u5f97 \u4ee5 \u4f7f \u7528 \u5c01 \u9589 \u89e3 \u7684 \u65b9 \u5f0f \u9032 \u884c \u53c3 \u6578 \u5167 \u5bb9 \u4e4b \u66f4 \u65b0 \u3002 3.2 \u805a \u96c6 \u4e8b \u5f8c \u6a5f \u7387 \u7dda \u6027 \u8ff4 \u6b78 (AAPLR)\u53c3\u6578\u4f30\u6e2c \u63a5 \u4e0b \u4f86 \u5c07 \u5229 \u7528 (27)\u5f0f\u9032\u884c\u6a21\u578b\u53c3\u6578\u7684\u4f30\u6e2c\uff0c\u8207\u4e00\u822c\u5316\u6700\u5c0f\u932f\u8aa4\u7387\u4e00\u6a23\uff0c\u540c\u6a23\u53ef\u5c07(27)\u5f0f\u6539\u5beb\u70ba\u4e00\u76ee\u6a19\u51fd\u5f0f\u70ba , , 1 1 1 1 1 1 ( ) m m N N M M m n m n m n m n J l d l M M = = = = = = \u2211\u2211 \u2211\u2211 % (35) , ,", |
| "eq_num": "( ) , ( ) log" |
| } |
| ], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "EQUATION", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [ |
| { |
| "start": 0, |
| "end": 8, |
| "text": "EQUATION", |
| "ref_id": "EQREF", |
| "raw_str": "( | ) ( ) log ( | ) ( ) m n m n m m r m m n j j r m j m d p Pg p Pg \u03bb \u03bb \u2260 = \u2212 \u2211 x W x W (36) \u5176 \u4e2d \uff0c ) ( ) (m r g W \u70ba \u8f49 \u63db \u77e9 \u9663 ) (m r W \u7684 \u4e8b \u524d \u6a5f \u7387 \u5206 \u4f48 \uff0c ) (m r \u4ee3 \u8868 \u6a21 \u578b m \u7684 \u8ff4 \u6b78 \u985e \u5225 \uff0c ) ( ) (m r g W \u70ba \u4e00 \u77e9 \u9663 \u7248 \u672c \u9ad8 \u65af \u5206 \u4f48 \uff0c \u7a31 \u4f5c elliptically symmetric distribution \u6216 matrix variate normal distribution\u3002 1 1 2 ( ) ( ) ( ) ( ) ( ) 1 ( ) ( ) ( ) D T r m r m d r m d d r m d r m d d g q \u2212 \u2212 = \u239b \u239e \u221d \u2206 \u22c5 \u2212 \u2212 \u239c \u239f \u239d \u23a0 \u2211 W w m \u03a3 w m (37) q \u70ba \u4e00 \u500b [ ) \u221e , 0 \u7684 \u51fd \u5f0f \uff0c d m r ) ( w \u548c d m r ) ( m \u5206 \u5225 \u4ee3 \u8868 \u8f49 \u63db \u77e9 \u9663 \u548c \u5e73 \u5747 \u77e9 \u9663 \u7684 \u7b2c d \u5217 \u5411 \u91cf \uff0c \u7dad \u5ea6 \u70ba ) 1 ( 1 + \u00d7 D \uff0c \u2206 \u70ba \u4e00 \u7dad \u5ea6 ) 1 ( ) 1 ( + \u00d7 + D D D D \u7684 \u5340 \u584a \u5c0d \u89d2 \u5316 \u5171 \u8b8a \u7570 \u77e9 \u9663 (block diagonal covariance matrix) \uff0c \u6bcf \u4e00 \u5340 \u584a \u7531 ) 1 ( ) 1 ( + \u00d7 + D D \u7684 d \u03a3 \u7d44 \u6210 \u3002 \u70ba \u4e86 \u7c21 \u5316 \u6700 \u4f73 \u8f49 \u63db \u77e9 \u9663 \uff0c \u9996 \u5148 \u5c07 \u52a0 \u5165 \u8f49 \u63db \u77e9 \u9663 \u7684 \u9ad8 \u65af \u5206 \u4f48 \u6539 \u5beb \u70ba \u4e00 \u55ae \u8b8a \u91cf \u5f62 \u5f0f \u5982 \u4e0b ( ) 2 , ,", |
| "eq_num": "( ) , , , , 1 2 2" |
| } |
| ], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "EQUATION", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [ |
| { |
| "start": 0, |
| "end": 8, |
| "text": "EQUATION", |
| "ref_id": "EQREF", |
| "raw_str": "2 1 , , , ( ) 1 1 ( | , ) exp 2 2 D m n d r m d m i m n m i m i D d m i d m i N \u03be \u03be \u03c3 \u03c0 = \u23a1 \u23a4 \u2212 = \u2212 \u23a2 \u23a5 \u23a2 \u23a5 \u23a3 \u23a6 \u2211 x w x \u03a3 \u03a3 (38) d m r ) ( w \u4ee3 \u8868 \u8f49 \u63db \u77e9 \u9663 ) (m r W \u7684 \u7b2c d \u5217 \u5411 \u91cf \u3002 \u5c07 (36)\u548c\u8b8a\u66f4\u904e\u7684\u9ad8\u65af\u5206\u4f48(38)\u4ee3\u5165(35)\u5f0f\u5f97\u5230 AAPLR \u7684 \u76ee \u6a19 \u51fd \u5f0f \u4e26 \u5c0d \u6b32 \u6c42 \u7684 \u8f49 \u63db \u77e9 \u9663 \u7b2c d \u5217 d m r ) ( w ( ) D d , , 1 K = \u53d6 \u504f \u5fae \u5206 \u5f97 ( ) , , 1 1 , ,", |
| "eq_num": "( ) , , ( ) , , 2 1 , () , , 1 ( ) ( ) ( ) , ' ( ) ( ) ' , ( ) (" |
| } |
| ], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "EQUATION", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [ |
| { |
| "start": 0, |
| "end": 8, |
| "text": "EQUATION", |
| "ref_id": "EQREF", |
| "raw_str": "\u2207 = \u2212 \u239b \u239e \u2212 \u239c \u239f \u239c \u239f \u239d \u23a0 + \u2212 \u2212 \u00d7 \u2211\u2211 \u2211 \u2211 w x W x w x W w m \u03a3 x W W x W W ( ) ' ' ', ,", |
| "eq_num": "' ( ) , , ( ) ', ', 2 , ' ( ) ' ,, 1 1 ( ) ( ) ( ) ( | , ) ( | , ) 2(" |
| } |
| ], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "EQUATION", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [ |
| { |
| "start": 0, |
| "end": 8, |
| "text": "EQUATION", |
| "ref_id": "EQREF", |
| "raw_str": "= \u2212 \u23a1 \u23a4 \u23a2 \u23a5 \u23a2 \u23a5 \u23a2 \u23a5 \u23a2 \u23a5 \u23a2 \u23a5 \u23a2 \u23a5 \u23a2 \u23a5 \u23a2 \u23a5 \u23a2 \u23a5 \u23a1 \u23a4 \u239b \u239e \u2212 \u23a2 \u23a5 \u23a2 \u23a5 \u239c \u239f \u23a2 \u23a5 \u239c \u239f \u00d7 \u23a2 \u23a5 \u239d \u23a0 \u23a2 \u23a5 \u23a2 \u23a5 \u23a2 \u23a5 + \u2212 \u23a2 \u23a5 \u23a3 \u23a6 \u23a3 \u23a6 \u2211 \u2211 W x W x w x W w m \u03a3 (39) 1 ( )(1 ( )) ( ) 2 ( )( 1 ( )) 1 ( )(1 ( )) ( ) ( ) 1 , ,", |
| "eq_num": "( ) ( ) 1" |
| } |
| ], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "EQUATION", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [ |
| { |
| "start": 0, |
| "end": 8, |
| "text": "EQUATION", |
| "ref_id": "EQREF", |
| "raw_str": "= = \u239b \u239e \u239c \u239f \u239c \u239f \u239c \u239f \u239c \u239f \u239c \u239f \u239c \u239f \u239c \u239f \u239c \u239f \u239c \u239f \u239c \u239f \u239c \u239f + \u2212 \u03a3 \u239c \u239f \u239d \u23a0 = \u2212 \u2126 \u2212 \u2212 \u03a3 \u2211 \u2211\u2211 \u2211 \u2211\u2211\u2211 \u2211 W x x x m ' , ,", |
| "eq_num": ", , ', , , ', , ', 2" |
| } |
| ], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "EQUATION", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [ |
| { |
| "start": 0, |
| "end": 8, |
| "text": "EQUATION", |
| "ref_id": "EQREF", |
| "raw_str": "1 1 ' 1 ', , ' 1 , , ', ,", |
| "eq_num": ", ( ) ( )" |
| } |
| ], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "EQUATION", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [ |
| { |
| "start": 0, |
| "end": 8, |
| "text": "EQUATION", |
| "ref_id": "EQREF", |
| "raw_str": "= = \u2208 \u2260 \u239b \u239e \u239c \u239f \u239c \u239f \u239c \u239f \u239c \u239f \u239c \u239f \u239c \u239f \u239c \u239f + \u2212 \u2126 \u239c \u239f \u239c \u239f \u239c \u239f \u239c \u239f + \u2212 \u03a3 \u239c \u239f \u239d \u23a0 \u2211 \u2211\u2211 \u2211 \u2211 \u2211\u2211 \u2211 W W x x x x m (40) \u5176 \u4e2d , ,", |
| "eq_num": "() , , , () (" |
| } |
| ], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "EQUATION", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [ |
| { |
| "start": 0, |
| "end": 8, |
| "text": "EQUATION", |
| "ref_id": "EQREF", |
| "raw_str": "\u2126 = x W x x W (41) , ' ( ) ( ) ', ,", |
| "eq_num": ", , ( ) (" |
| } |
| ], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "\u4ee4 (40)\u5f0f\u7684\u7b49\u865f\u5de6\u5074\u70ba L w \u22c5 d m r ) ( \uff0c L \u70ba \u4e00 \u7dad \u5ea6 ) 1 ( ) 1 ( + \u00d7 + D D \u7684 \u65b9 \u9663 \uff0c \u4e14 \u4ee4 \u7b49 \u865f \u53f3 \u5074 \u70ba r\uff0c\u7dad\u5ea6 ) 1 ( 1 + \u00d7 D \uff0c (40)\u5f0f\u8b8a\u70ba r L w = \u22c5 d m r ) ( \uff0c \u53ef \u5f97 \u8f49 \u63db \u77e9 \u9663 ) (m r W \u7684 \u7b2c d \u5217 \u70ba 1 ) ( \u2212 \u22c5 = L r w d m r (43) \u91cd \u8986 \u4e0a \u8ff0 \u6b65 \u9a5f \uff0c \u53ef \u6c42 \u5f97 \u6240 \u6709 \u8ff4 \u6b78 \u985e \u5225 \u7684 \u8f49 \u63db \u77e9 \u9663 \u3002 4. \u5be6\u9a57\u8207\u8a0e\u8ad6 4.1 \u5be6 \u9a57 \u74b0 \u5883 \u8207 \u8a9e \u97f3 \u53c3 \u6578 \u3001 \u6a21 \u578b \u8a2d \u5b9a \u5728 \u5be6 \u9a57 \u7684 \u786c \u9ad4 \u65b9 \u9762 \uff0c \u6211 \u5011 \u6240 \u4f7f \u7528 \u7684 \u662f Pentium 4 2.0GHz \u500b \u4eba \u96fb \u8166 \uff0c \u642d \u914d 256MB \u7684 \u8a18 \u61b6 \u9ad4 \u5bb9 \u91cf \uff0c \u4f7f \u7528 \u7684 \u4f5c \u696d \u7cfb \u7d71 \u70ba Windows XP Professional \u4e2d \u6587 \u7248 \uff0c \u4e26 \u4ee5 Microsoft Visual C++ 6.0 \u4f5c \u70ba \u8edf \u9ad4 \u958b \u767c \u5de5 \u5177 \u3002 \u5728 \u8a9e \u97f3 \u7279 \u5fb5 \u53c3 \u6578 \u6c42 \u53d6 \u90e8 \u4efd \uff0c \u4f7f \u7528 HTK \u4e2d \u7684 HCopy \u6307 \u4ee4 \u53d6 \u51fa \u8a9e \u97f3 \u7279 \u5fb5 \u53c3 \u6578 \uff0c \u6bcf \u4e00 \u97f3 \u6846 \u7684 \u7279 \u5fb5 \u53c3 \u6578 \u7686 \u70ba 26 \u7dad \uff0c \u5176 \u4e2d \u5305 \u62ec 12 \u968e \u7684 MFCC\uff0c12 \u968e \u7684 delta MFCC\uff0c1 \u968e log energy \u4ee5 \u53ca 1 \u968e delta log energy\uff0c\u95dc\u65bc", |
| "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": "\u4e4b \u57fa \u672c \u55ae \u5143 \uff0c \u5728 \u8072 \u6bcd \u90e8 \u4efd \u4f7f \u7528 3 \u500b \u72c0 \u614b \u8868 \u793a \uff0c \u800c \u97fb \u6bcd \u5247 \u4f7f \u7528 5 \u500b \u72c0 \u614b \u4f86 \u8868 \u793a \u3002 \u540c \u6642 \uff0c \u6df7 \u5408 \u6578 \u6578 \u91cf \u5247 \u6839 \u64da \u5404 \u500b \u72c0 \u614b \u6240 \u5206 \u914d \u5230 \u7684 \u97f3 \u6846 \u6578 \u91cf \u4f86 \u6c7a \u5b9a \uff0c \u4f46 \u6700 \u5927 \u6df7 \u5408 \u6578 \u4e0d \u5f97 \u8d85 \u904e 32 \u500b \u3002 4.2 \u5be6 \u9a57 \u8a9e \u6599 \u5728 \u672c \u5be6 \u9a57 \u4e2d \uff0c \u6211 \u5011 \u5206 \u5225 \u4f7f \u7528 \u5169 \u7a2e \u8a9e \u6599 \u4ee5 \u9032 \u884c \u5be6 \u9a57 \uff0c \u5176 \u4e00 \u662f TCC300 \u9ea5 \u514b \u98a8 \u8a9e \u6599 \u5eab \uff0c \u7528 \u65bc \u8a9e \u8005 \u7368 \u7acb \u4e4b \u8a9e \u97f3 \u6a21 \u578b \u53c3 \u6578 \u8a13 \u7df4 \uff1b \u53e6 \u4e00 \u500b \u5247 \u662f \u516c \u8996 \u665a \u9593 \u65b0 \u805e \u8a9e \u6599 \uff0c \u7528 \u65bc \u8abf \u9069 \u5f8c \u4e4b \u8fa8 \u8b58 \u6548 \u80fd \u6539 \u9032 \u8a55 \u4f30 \u3002 \u4ee5 \u4e0b \u662f \u9019 \u5169 \u500b \u8a9e \u6599 \u5eab \u7684 \u8cc7 \u6599 \u8aaa \u660e \u3002 TCC \u53f0 \u5927 /\u6210\u5927/\u4ea4\u5927\u9ea5\u514b\u98a8\u8a9e\u97f3\u8cc7\u6599\u5eab\u662f\u7531\u570b\u7acb\u53f0\u7063\u5927\u5b78\u3001\u570b\u7acb\u6210\u529f\u5927\u5b78\u3001\u570b\u7acb\u4ea4\u901a\u5927\u5b78\u5404\u81ea\u64c1\u6709\u4e4b\u8a9e\u6599\u5eab\u96c6 \u5408 \u800c \u6210 \uff0c \u5404 \u6821 \u9304 \u88fd \u4e4b \u76ee \u7684 \u662f \u70ba \u8a9e \u97f3 \u8fa8 \u8a8d \u7814 \u7a76 \uff0c \u5c6c \u65bc \u9ea5 \u514b \u98a8 \u6717 \u8b80 \u8a9e \u97f3 \u3002 \u5176 \u4e2d \u53f0 \u5927 \u8a9e \u6599 \u5eab \u4e3b \u8981 \u5305 \u542b \u8a5e \u53ca \u77ed \u53e5 \uff0c \u5167 \u5bb9 \u7d93 \u904e \u4ed4 \u7d30 \u8a2d \u8a08 \uff0c \u8003 \u616e \u4e86 \u97f3 \u7bc0 \u53ca \u5176 \u76f8 \u9023 \u51fa \u73fe \u6a5f \u7387 \uff0c \u7531 100 \u4eba \u9304 \u88fd \u800c \u6210 \uff1b \u6210 \u5927 \u53ca \u4ea4 \u5927 \u8a9e \u6599 \u5eab \u4e3b \u8981 \u5305 \u542b \u9577 \u6587 \u8a9e \u6599 \uff0c \u6587 \u7ae0 \u7531 \u4e2d \u7814 \u9662 \u63d0 \u4f9b \u4e4b 500 \u842c \u8a5e \u8a5e \u985e \u6a19 \u793a \u8a9e \u6599 \u5eab \u4e2d \u9078 \u53d6 \uff0c \u6bcf \u7bc7 \u6587 \u7ae0 \u5305 \u542b \u6578 \u767e \u5b57 \uff0c \u518d \u5207 \u5272 \u6210 3-4 \u6bb5 \uff0c \u6bcf \u6bb5 \u542b \u81f3 \u591a 231 \u5b57 \uff0c \u7531 200 \u4eba \u6717 \u8b80 \u9304 \u88fd \uff0c \u6bcf \u4eba \u6240 \u8b80 \u6587 \u7ae0 \u7686 \u4e0d \u76f8 \u540c \u3002 \u5728 \u672c \u8ad6 \u6587 \u7684 \u5be6 \u9a57 \u4e2d \uff0c \u6211 \u5011 \u6240 \u53d6 \u7684 \u90e8 \u4efd \u70ba \u4ea4 \u901a \u5927 \u5b78 \u53ca \u53f0 \u7063 \u5927 \u5b78 \u6240 \u9304 \u88fd \u7684 \u97f3 \u6a94 \u3002 \u8a9e \u97f3 \u8a0a \u865f \u53d6 \u6a23 \u983b \u7387 \u70ba 16kHz\uff0c\u8a9e\u97f3\u8a0a\u865f\u91cf\u5316\u6790\u5ea6\u70ba 16 \u4f4d \u5143 \u3002 \u516c \u8996 \u665a \u9593 \u65b0 \u805e \u8a9e \u6599 \u662f \u7531 \u4e2d \u592e \u7814 \u7a76 \u9662 \u8207 \u516c \u5171 \u96fb \u8996 \u81fa \u5171 \u540c \u9304 \u88fd \uff0c \u4e3b \u8981 \u662f \u516c \u8996 \u665a \u9593 \u65b0 \u805e \u8a9e \u97f3 \uff0c \u9304 \u88fd \u671f \u9593 \u7531 2000 \u5e74 1 \u6708 11 \u65e5 \u5230 2000 \u5e74 2 \u6708 9 \u65e5 \uff0c \u7e3d \u5171 120 \u5c0f \u6642 \u7684 \u65b0 \u805e \u8a9e \u6599 \u3002 4.3 \u5be6 \u9a57 \u7d50 \u679c \u8207 \u8a0e \u8ad6 \u9996 \u5148 \uff0c \u6211 \u5011 \u4f7f \u7528 TCC300 \u8a9e \u6599 \u5eab \u9032 \u884c \u8a9e \u8005 \u7368 \u7acb \u4e4b \u8a9e \u97f3 \u6a21 \u578b \u8a13 \u7df4 \uff0c \u8a9e \u6599 \u6578 \u5171 \u7d04 14000 \u53e5 \u3002 \u8a13 \u7df4 \u6240 \u5f97 \u4e4b \u8a9e \u97f3 \u6a21 \u578b \uff0c \u6211 \u5011 \u4f7f \u7528 TCC300 \u4e2d \u53e6 \u5916 \u672a \u62ff \u4f86 \u8a13 \u7df4 \u4e4b \u8a9e \u6599 \u9032 \u884c \u6e2c \u8a66 \u5171 900 \u53e5 \uff0c \u5176 \u8a9e \u97f3 \u8fa8 \u8b58 \u7387 \u70ba 67.5%\u3002\u53e6\u4e00\u4efd\u6211\u5011\u6240\u4f7f\u7528\u4e4b\u516c\u8996 \u665a \u9593 \u65b0 \u805e \u8a9e \u6599 \uff0c \u521d \u6b65 \u5df2 \u6574 \u7406 \u51fa \u4e09 \u5c0f \u6642 \u8a9e \u6599 \u3002 \u6240 \u4ee5 \u5728 \u63a5 \u4e0b \u4f86 \u7684 \u6548 \u80fd \u6e2c \u8a66 \u90e8 \u4efd \uff0c \u6211 \u5011 \u4f7f \u7528 \u6b64 \u4e09 \u5c0f \u6642 \u8a9e \u6599 \u9032 \u884c \u5be6 \u9a57 \u3002 \u6211 \u5011 \u5c07 \u4f7f \u7528 \u4e0d \u540c \u7684 \u8abf \u9069 \u8a9e \u6599 \u91cf \uff0c \u5206 \u5225 \u9032 \u884c \u6700 \u5927 \u76f8 \u4f3c \u5ea6 \u7dda \u6027 \u8ff4 \u6b78 (MLLR)\u3001\u6700\u5927\u4e8b\u5f8c\u6a5f\u7387\u7dda\u6027\u8ff4\u6b78(MAPLR)\u3001\u6700\u5c0f\u5206\u985e\u932f\u8aa4 \u7dda \u6027 \u8ff4 \u6b78 (MCELR)\u8207\u672c\u8ad6\u6587\u6240\u63d0\u4e4b\u805a\u96c6\u4e8b\u5f8c\u6a5f\u7387\u7dda\u6027\u8ff4\u6b78(AAPLR)\u4e4b\u6548\u80fd\u8a55\u4f30\u3002\u8abf\u9069\u4e4b\u8a9e\u6599\u91cf\u7531\u6700\u77ed\u4e4b 2 \u53e5 \uff0c \u81f3 \u6700 \u9577 \u4e4b 30 \u53e5 \uff0c \u800c \u8abf \u9069 \u4e4b \u8f49 \u63db \u77e9 \u9663 \u985e \u5225 \u70ba 2 \u985e \uff0c \u9032 \u884c \u6548 \u80fd \u4e4b \u8a55 \u4f30 \uff0c \u5be6 \u9a57 \u7d50 \u679c \u5982 \u4e0b \u8868 \u6240 \u793a \u3002 \u8abf \u9069 \u65b9 \u6cd5 \u8abf \u9069 \u53e5 \u6578 \u77e9 \u9663 \u985e \u5225 \u6578 \u8fa8 \u8b58 \u7387 (%) \u8abf \u9069 \u6642 \u9593", |
| "eq_num": "(" |
| } |
| ], |
| "section": "", |
| "sec_num": null |
| } |
| ], |
| "back_matter": [ |
| { |
| "text": " 1 1, ,", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "annex", |
| "sec_num": null |
| } |
| ], |
| "bib_entries": { |
| "BIBREF0": { |
| "ref_id": "b0", |
| "title": "Maximum mutual information estimation of hidden Markov model parameters for speech recognition", |
| "authors": [ |
| { |
| "first": "L", |
| "middle": [], |
| "last": "Bahl", |
| "suffix": "" |
| }, |
| { |
| "first": "P", |
| "middle": [], |
| "last": "Brown", |
| "suffix": "" |
| }, |
| { |
| "first": "P", |
| "middle": [], |
| "last": "Souza", |
| "suffix": "" |
| }, |
| { |
| "first": "R", |
| "middle": [], |
| "last": "Mercer", |
| "suffix": "" |
| } |
| ], |
| "year": 1986, |
| "venue": "Proc. IEEE Int. Conf. Acoustics, Speech, Signal Processing (ICASSP)", |
| "volume": "11", |
| "issue": "", |
| "pages": "49--52", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "L. Bahl, P. Brown, P. de Souza and R. Mercer, \"Maximum mutual information estimation of hidden Markov model parameters for speech recognition\", in Proc. IEEE Int. Conf. Acoustics, Speech, Signal Processing (ICASSP), vol. 11, April 1986, pp. 49-52.", |
| "links": null |
| }, |
| "BIBREF1": { |
| "ref_id": "b1", |
| "title": "Discriminative model combination", |
| "authors": [ |
| { |
| "first": "P", |
| "middle": [], |
| "last": "Beyerlin", |
| "suffix": "" |
| } |
| ], |
| "year": 1998, |
| "venue": "Proc. IEEE Int. Conf. Acoustics, Speech, Signal Processing (ICASSP)", |
| "volume": "1", |
| "issue": "", |
| "pages": "481--485", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "P. Beyerlin, \"Discriminative model combination\", in Proc. IEEE Int. Conf. Acoustics, Speech, Signal Processing (ICASSP), vol. 1, 1998, pp. 481-485.", |
| "links": null |
| }, |
| "BIBREF2": { |
| "ref_id": "b2", |
| "title": "Discriminative training of dynamic programming based speech recognizers", |
| "authors": [ |
| { |
| "first": "P", |
| "middle": [ |
| "C" |
| ], |
| "last": "Chang", |
| "suffix": "" |
| }, |
| { |
| "first": "B.-H", |
| "middle": [], |
| "last": "Juang", |
| "suffix": "" |
| } |
| ], |
| "year": 1993, |
| "venue": "IEEE Trans. Speech and Audio Processing", |
| "volume": "1", |
| "issue": "2", |
| "pages": "135--143", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "P. C. Chang and B.-H. Juang, \"Discriminative training of dynamic programming based speech recognizers\", IEEE Trans. Speech and Audio Processing, vol. 1, no. 2, pp. 135-143, April 1993.", |
| "links": null |
| }, |
| "BIBREF3": { |
| "ref_id": "b3", |
| "title": "Speaker adaptation using discriminative linear regression on time-varying mean parameters in trended HMM", |
| "authors": [ |
| { |
| "first": "R", |
| "middle": [], |
| "last": "Chengalvarayan", |
| "suffix": "" |
| } |
| ], |
| "year": 1998, |
| "venue": "IEEE Trans. Signal Processing Letters", |
| "volume": "5", |
| "issue": "", |
| "pages": "63--65", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "R. Chengalvarayan, \"Speaker adaptation using discriminative linear regression on time-varying mean parameters in trended HMM\", IEEE Trans. Signal Processing Letters, vol. 5, pp. 63-65, March 1998.", |
| "links": null |
| }, |
| "BIBREF4": { |
| "ref_id": "b4", |
| "title": "Quasi-Bayes linear regression for sequential learning of hidden Markov models", |
| "authors": [ |
| { |
| "first": "Jen-Tzung", |
| "middle": [], |
| "last": "Chien", |
| "suffix": "" |
| } |
| ], |
| "year": 2002, |
| "venue": "IEEE Trans. on Speech and Audio Processing", |
| "volume": "10", |
| "issue": "5", |
| "pages": "268--278", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Jen-Tzung Chien, \"Quasi-Bayes linear regression for sequential learning of hidden Markov models\", IEEE Trans. on Speech and Audio Processing, vol. 10, no. 5, pp. 268-278, July 2002.", |
| "links": null |
| }, |
| "BIBREF5": { |
| "ref_id": "b5", |
| "title": "Maximum likelihood from incomplete data via the EM algorithm", |
| "authors": [ |
| { |
| "first": "A", |
| "middle": [ |
| "P" |
| ], |
| "last": "Dempster", |
| "suffix": "" |
| }, |
| { |
| "first": "N", |
| "middle": [ |
| "M" |
| ], |
| "last": "Laird", |
| "suffix": "" |
| }, |
| { |
| "first": "D", |
| "middle": [ |
| "B" |
| ], |
| "last": "Rubin", |
| "suffix": "" |
| } |
| ], |
| "year": 1977, |
| "venue": "Journal of the Royal Statistical Society (B)", |
| "volume": "39", |
| "issue": "", |
| "pages": "1--38", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "A. P. Dempster, N. M. Laird and D. B. Rubin, \"Maximum likelihood from incomplete data via the EM algorithm\", Journal of the Royal Statistical Society (B), vol. 39, pp. 1-38, 1977.", |
| "links": null |
| }, |
| "BIBREF6": { |
| "ref_id": "b6", |
| "title": "Mean and Variance adaptation within the MLLR Framework", |
| "authors": [ |
| { |
| "first": "M", |
| "middle": [ |
| "J F" |
| ], |
| "last": "Gales", |
| "suffix": "" |
| }, |
| { |
| "first": "P", |
| "middle": [ |
| "C" |
| ], |
| "last": "Woodland", |
| "suffix": "" |
| } |
| ], |
| "year": 1996, |
| "venue": "Computer Speech and Language", |
| "volume": "10", |
| "issue": "", |
| "pages": "249--264", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "M. J. F. Gales and P. C. Woodland, \"Mean and Variance adaptation within the MLLR Framework,\" Computer Speech and Language, Vol. 10, pp. 249-264, 1996.", |
| "links": null |
| }, |
| "BIBREF7": { |
| "ref_id": "b7", |
| "title": "Maximum a posteriori estimation for multivariate Gaussian mixture observation of Markov chains", |
| "authors": [ |
| { |
| "first": "J.-L", |
| "middle": [], |
| "last": "Gauvain", |
| "suffix": "" |
| }, |
| { |
| "first": "C.-H", |
| "middle": [], |
| "last": "Lee", |
| "suffix": "" |
| } |
| ], |
| "year": 1994, |
| "venue": "IEEE Transactions on Speech and Audio Processing", |
| "volume": "2", |
| "issue": "4", |
| "pages": "291--298", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "J.-L. Gauvain and C.-H. Lee, \"Maximum a posteriori estimation for multivariate Gaussian mixture observation of Markov chains\", IEEE Transactions on Speech and Audio Processing, vol. 2, no. 4, pp. 291-298, April 1994.", |
| "links": null |
| }, |
| "BIBREF8": { |
| "ref_id": "b8", |
| "title": "Minimum classification error linear regression for acoustic model adaptation of continuous density HMMs", |
| "authors": [ |
| { |
| "first": "X", |
| "middle": [], |
| "last": "He", |
| "suffix": "" |
| }, |
| { |
| "first": "W", |
| "middle": [], |
| "last": "Chou", |
| "suffix": "" |
| } |
| ], |
| "year": 2003, |
| "venue": "Proc. Int. Conf. Multimedia and Expo (ICME)", |
| "volume": "1", |
| "issue": "", |
| "pages": "6--9", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "X. He, W. Chou, \"Minimum classification error linear regression for acoustic model adaptation of continuous density HMMs\", in Proc. Int. Conf. Multimedia and Expo (ICME), vol. 1, 2003, pp. 6-9.", |
| "links": null |
| }, |
| "BIBREF9": { |
| "ref_id": "b9", |
| "title": "Supervised adaptation of MCE-trained CDHMMs using minimum classification error linear regression", |
| "authors": [ |
| { |
| "first": "W", |
| "middle": [], |
| "last": "Jian", |
| "suffix": "" |
| }, |
| { |
| "first": "H", |
| "middle": [], |
| "last": "Qiang", |
| "suffix": "" |
| } |
| ], |
| "year": 2002, |
| "venue": "Proc. IEEE Int. Conf. Acoustic, Speech, Signal Processing (ICASSP)", |
| "volume": "1", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "W. Jian, H. Qiang, \"Supervised adaptation of MCE-trained CDHMMs using minimum classification error linear regression,\" in Proc. IEEE Int. Conf. Acoustic, Speech, Signal Processing (ICASSP), vol. 1, 2002, pp. I-605 -I-608.", |
| "links": null |
| }, |
| "BIBREF10": { |
| "ref_id": "b10", |
| "title": "Minimum classification error rate Methods for Speech Recognition", |
| "authors": [ |
| { |
| "first": "B.-H", |
| "middle": [], |
| "last": "Juang", |
| "suffix": "" |
| }, |
| { |
| "first": "W", |
| "middle": [], |
| "last": "Hou", |
| "suffix": "" |
| }, |
| { |
| "first": "C.-H", |
| "middle": [], |
| "last": "Lee", |
| "suffix": "" |
| } |
| ], |
| "year": 1997, |
| "venue": "IEEE Trans. Speech and Audio Processing", |
| "volume": "5", |
| "issue": "3", |
| "pages": "257--265", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "B.-H. Juang, W. Hou and C.-H. Lee, \"Minimum classification error rate Methods for Speech Recognition\", IEEE Trans. Speech and Audio Processing, vol. 5, no. 3 , pp. 257-265, May 1997.", |
| "links": null |
| }, |
| "BIBREF11": { |
| "ref_id": "b11", |
| "title": "Discriminative learning for minimum error classification", |
| "authors": [ |
| { |
| "first": "B.-H", |
| "middle": [], |
| "last": "Juang", |
| "suffix": "" |
| }, |
| { |
| "first": "S", |
| "middle": [], |
| "last": "Katagirl", |
| "suffix": "" |
| } |
| ], |
| "year": 1992, |
| "venue": "IEEE Trans. Signal Processing", |
| "volume": "40", |
| "issue": "", |
| "pages": "3043--3054", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "B.-H. Juang and S. Katagirl, \"Discriminative learning for minimum error classification,\" IEEE Trans. Signal Processing, vol. 40, pp. 3043-3054, December 1992.", |
| "links": null |
| }, |
| "BIBREF12": { |
| "ref_id": "b12", |
| "title": "Discriminative training of language models for speech recognition", |
| "authors": [ |
| { |
| "first": "H.-K", |
| "middle": [ |
| "J" |
| ], |
| "last": "Kuo", |
| "suffix": "" |
| }, |
| { |
| "first": "E", |
| "middle": [], |
| "last": "Fosle-Lussier", |
| "suffix": "" |
| }, |
| { |
| "first": "H", |
| "middle": [], |
| "last": "Jiang", |
| "suffix": "" |
| }, |
| { |
| "first": "C.-H", |
| "middle": [], |
| "last": "Lee", |
| "suffix": "" |
| } |
| ], |
| "year": 2002, |
| "venue": "Proc. IEEE Int. Conf. Acoustics, Speech, Signal Processing (ICASSP)", |
| "volume": "1", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "H.-K.J. Kuo, E. Fosle-Lussier, H. Jiang and C.-H. Lee, \"Discriminative training of language models for speech recognition\", in Proc. IEEE Int. Conf. Acoustics, Speech, Signal Processing (ICASSP), vol. 1, 2002, pp. I-325-I-328.", |
| "links": null |
| }, |
| "BIBREF13": { |
| "ref_id": "b13", |
| "title": "Maximum Likelihood Linear Regression for Speaker Adaptation of Continuous Density Hidden Markov Models", |
| "authors": [ |
| { |
| "first": "C", |
| "middle": [ |
| "J" |
| ], |
| "last": "Leggeter", |
| "suffix": "" |
| }, |
| { |
| "first": "P", |
| "middle": [ |
| "C" |
| ], |
| "last": "Woodland", |
| "suffix": "" |
| } |
| ], |
| "year": 1995, |
| "venue": "Computer Speech and Language", |
| "volume": "", |
| "issue": "", |
| "pages": "171--185", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "C. J. Leggeter and P.C. Woodland, \"Maximum Likelihood Linear Regression for Speaker Adaptation of Continuous Density Hidden Markov Models,\" Computer Speech and Language, 1995, pp. 171-185.", |
| "links": null |
| }, |
| "BIBREF14": { |
| "ref_id": "b14", |
| "title": "A new algorithm for fast discriminative training", |
| "authors": [ |
| { |
| "first": "Q", |
| "middle": [], |
| "last": "Li", |
| "suffix": "" |
| }, |
| { |
| "first": "B.-H", |
| "middle": [], |
| "last": "Juang", |
| "suffix": "" |
| } |
| ], |
| "year": 2002, |
| "venue": "Proc. Int. Conf. Acoustics, Speech, Signal Processing (ICASSP)", |
| "volume": "1", |
| "issue": "", |
| "pages": "97--100", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Q. Li, B.-H. Juang, \"A new algorithm for fast discriminative training\", in Proc. Int. Conf. Acoustics, Speech, Signal Processing (ICASSP), vol. 1, 2002, pp. 97-100.", |
| "links": null |
| }, |
| "BIBREF15": { |
| "ref_id": "b15", |
| "title": "Fast discriminative training for sequential observations with application to speaker identification", |
| "authors": [ |
| { |
| "first": "Q", |
| "middle": [], |
| "last": "Li", |
| "suffix": "" |
| }, |
| { |
| "first": "B.-H", |
| "middle": [], |
| "last": "Juang", |
| "suffix": "" |
| } |
| ], |
| "year": 2003, |
| "venue": "Proc. Int. Conf. Acoustics, Speech, Signal Processing (ICASSP)", |
| "volume": "2", |
| "issue": "", |
| "pages": "397--400", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Q. Li, B.-H. Juang, \"Fast discriminative training for sequential observations with application to speaker identification\", in Proc. Int. Conf. Acoustics, Speech, Signal Processing (ICASSP), vol. 2, 2003, pp. 397-400.", |
| "links": null |
| }, |
| "BIBREF16": { |
| "ref_id": "b16", |
| "title": "An intorduction to computing with neural nets", |
| "authors": [ |
| { |
| "first": "R", |
| "middle": [ |
| "P" |
| ], |
| "last": "Lippmann", |
| "suffix": "" |
| } |
| ], |
| "year": 1987, |
| "venue": "IEEE ASSP Mag", |
| "volume": "", |
| "issue": "", |
| "pages": "4--22", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "R. P. Lippmann, \"An intorduction to computing with neural nets\", IEEE ASSP Mag., pp. 4-22, April 1987.", |
| "links": null |
| }, |
| "BIBREF17": { |
| "ref_id": "b17", |
| "title": "Shift-invariant multi-category phoneme recognition using kohonen's LVQ2", |
| "authors": [ |
| { |
| "first": "E", |
| "middle": [], |
| "last": "Mcdermott", |
| "suffix": "" |
| }, |
| { |
| "first": "S", |
| "middle": [], |
| "last": "Katagiri", |
| "suffix": "" |
| } |
| ], |
| "year": 1989, |
| "venue": "Proc. IEEE Int. Conf. Acoustics, Speech, Signal Processing", |
| "volume": "", |
| "issue": "", |
| "pages": "81--84", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "E. McDermott and S. Katagiri, \"Shift-invariant multi-category phoneme recognition using kohonen's LVQ2,\" in Proc. IEEE Int. Conf. Acoustics, Speech, Signal Processing (ICASSP), 1989, pp. 81-84.", |
| "links": null |
| }, |
| "BIBREF18": { |
| "ref_id": "b18", |
| "title": "A maximum-likelihood approach to stochastic matching for robust speech recognition", |
| "authors": [ |
| { |
| "first": "A", |
| "middle": [], |
| "last": "Sankar", |
| "suffix": "" |
| }, |
| { |
| "first": "C.-H", |
| "middle": [], |
| "last": "Lee", |
| "suffix": "" |
| } |
| ], |
| "year": 1996, |
| "venue": "IEEE Trans. Speech and Audio Processing", |
| "volume": "4", |
| "issue": "3", |
| "pages": "190--202", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "A. Sankar and C.-H. Lee, \"A maximum-likelihood approach to stochastic matching for robust speech recognition\", IEEE Trans. Speech and Audio Processing, vol. 4, no. 3, pp. 190-202, May 1996.", |
| "links": null |
| }, |
| "BIBREF19": { |
| "ref_id": "b19", |
| "title": "A combined maximum mutual information and maximum likelihood approach for mixture density splitting", |
| "authors": [ |
| { |
| "first": "R", |
| "middle": [], |
| "last": "Schl\u00fcter", |
| "suffix": "" |
| }, |
| { |
| "first": "W", |
| "middle": [], |
| "last": "Macherey", |
| "suffix": "" |
| }, |
| { |
| "first": "B", |
| "middle": [], |
| "last": "M\u00fcller", |
| "suffix": "" |
| }, |
| { |
| "first": "H", |
| "middle": [], |
| "last": "Ney", |
| "suffix": "" |
| } |
| ], |
| "year": 1999, |
| "venue": "Proc. EUROSPEECH", |
| "volume": "4", |
| "issue": "", |
| "pages": "1715--1718", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "R. Schl\u00fcter, W. Macherey, B. M\u00fcller and H. Ney, \"A combined maximum mutual information and maximum likelihood approach for mixture density splitting\", in Proc. EUROSPEECH, vol. 4, 1999, pp. 1715-1718.", |
| "links": null |
| }, |
| "BIBREF20": { |
| "ref_id": "b20", |
| "title": "Hidden Markov model adaptation using maximum a posteriori linear regression", |
| "authors": [ |
| { |
| "first": "O", |
| "middle": [], |
| "last": "Siohan", |
| "suffix": "" |
| }, |
| { |
| "first": "C", |
| "middle": [], |
| "last": "Chesta", |
| "suffix": "" |
| }, |
| { |
| "first": "C.-H", |
| "middle": [], |
| "last": "Lee", |
| "suffix": "" |
| } |
| ], |
| "year": 1999, |
| "venue": "Workshop on Robust Methods for Speech Recognition in Adverse Conditions", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "O. Siohan, C. Chesta, and C.-H. Lee. \"Hidden Markov model adaptation using maximum a posteriori linear regression.\" in Workshop on Robust Methods for Speech Recognition in Adverse Conditions, Tampere, Finland, 1999.", |
| "links": null |
| }, |
| "BIBREF21": { |
| "ref_id": "b21", |
| "title": "The HTK Book (Version 2.0). ECRL", |
| "authors": [ |
| { |
| "first": "S", |
| "middle": [], |
| "last": "Young", |
| "suffix": "" |
| }, |
| { |
| "first": "J", |
| "middle": [], |
| "last": "Jansen", |
| "suffix": "" |
| }, |
| { |
| "first": "J", |
| "middle": [], |
| "last": "Odell", |
| "suffix": "" |
| }, |
| { |
| "first": "D", |
| "middle": [], |
| "last": "Ollason", |
| "suffix": "" |
| }, |
| { |
| "first": "P", |
| "middle": [], |
| "last": "Woodland", |
| "suffix": "" |
| } |
| ], |
| "year": 1995, |
| "venue": "", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "S. Young, J. Jansen, J. Odell, D. Ollason, and P Woodland. The HTK Book (Version 2.0). ECRL, 1995.", |
| "links": null |
| } |
| }, |
| "ref_entries": { |
| "FIGREF0": { |
| "num": null, |
| "text": "aggregate a posteriori, AAP)", |
| "type_str": "figure", |
| "uris": null |
| } |
| } |
| } |
| } |
