{ "paper_id": "O03-3008", "header": { "generated_with": "S2ORC 1.0.0", "date_generated": "2023-01-19T08:01:29.066927Z" }, "title": "", "authors": [], "year": "", "venue": null, "identifiers": {}, "abstract": "", "pdf_parse": { "paper_id": "O03-3008", "_pdf_hash": "", "abstract": [], "body_text": [], "back_matter": [], "bib_entries": { "BIBREF0": { "ref_id": "b0", "title": "MIRACLE: A Music Information Retrieval System with Clustered Computing Engines", "authors": [ { "first": "J.-S. Roger", "middle": [], "last": "Jang", "suffix": "" }, { "first": "Jiang-Chun", "middle": [], "last": "Chen", "suffix": "" }, { "first": "Ming-Yang", "middle": [], "last": "Kao", "suffix": "" } ], "year": 2001, "venue": "International Symposium on Music Information Retrieval", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "J.-S. Roger Jang, Jiang-Chun Chen, Ming-Yang Kao,\"MIRACLE: A Music Information Retrieval System with Clustered Computing Engines\", International Symposium on Music Information Retrieval (MUSIC IR 2001)", "links": null }, "BIBREF1": { "ref_id": "b1", "title": "Optimization of Viterbi Beam Search in Speech Recognition", "authors": [ { "first": "J. -S", "middle": [], "last": "Jang", "suffix": "" }, { "first": "", "middle": [], "last": "Roger", "suffix": "" }, { "first": "", "middle": [], "last": "Lin", "suffix": "" }, { "first": "", "middle": [], "last": "Shiuan-Sung", "suffix": "" } ], "year": 2002, "venue": "International Symposium on Chinese Spoken Language Processing", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "Jang, J. -S. Roger and Lin, Shiuan-Sung, \"Optimization of Viterbi Beam Search in Speech Recognition\", International Symposium on Chinese Spoken Language Processing, Taiwan, August 2002.", "links": null }, "BIBREF2": { "ref_id": "b2", "title": "Fundamentals of speech recognition", "authors": [ { "first": "Lawrence", "middle": [], "last": "Rabiner", "suffix": "" }, { "first": "B", "middle": [], "last": "Juang", "suffix": "" } ], "year": 1993, "venue": "", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "Lawrence Rabiner, B.H Juang, Fundamentals of speech recognition, Prentice Hall, 1993.", "links": null }, "BIBREF3": { "ref_id": "b3", "title": "Speech Communication : human and machine", "authors": [ { "first": "D", "middle": [], "last": "O\uff07shanughnessy", "suffix": "" } ], "year": 1987, "venue": "", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "O\uff07Shanughnessy, D., Speech Communication : human and machine, Addison-Wesley, 1987.", "links": null }, "BIBREF4": { "ref_id": "b4", "title": "Fundamentals of Speech Recognition", "authors": [ { "first": "L", "middle": [], "last": "Rabiner", "suffix": "" }, { "first": "B.-W", "middle": [], "last": "Juang", "suffix": "" } ], "year": 1993, "venue": "", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "Rabiner, L. and Juang, B.-W., Fundamentals of Speech Recognition. Prentice Hall PTR, Upper Saddle River, New Jersey, 1993", "links": null }, "BIBREF5": { "ref_id": "b5", "title": "The anatomy of a large-scale hypertextual Web search engine", "authors": [ { "first": "Sergey", "middle": [], "last": "Brin", "suffix": "" }, { "first": "Lawrence", "middle": [], "last": "Page", "suffix": "" } ], "year": null, "venue": "Ashman and Thistlewaite", "volume": "2", "issue": "", "pages": "107--117", "other_ids": {}, "num": null, "urls": [], "raw_text": "Sergey Brin and Lawrence Page. The anatomy of a large-scale hypertextual Web search engine. In Ashman and Thistlewaite [2], pages 107-117.Brisbane,Australia.http://citeseer.nj.nec.com/brin98anato my.html", "links": null }, "BIBREF6": { "ref_id": "b6", "title": "The HTK Book version 3, Microsoft Corporation", "authors": [ { "first": "Steven", "middle": [], "last": "Young", "suffix": "" } ], "year": 2000, "venue": "", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "Steven Young, The HTK Book version 3, Microsoft Corporation, 2000.", "links": null }, "BIBREF7": { "ref_id": "b7", "title": "Voice and Speech Processing", "authors": [ { "first": "T", "middle": [ "W" ], "last": "Parsons", "suffix": "" } ], "year": 1986, "venue": "", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "T.W. Parsons, Voice and Speech Processing, McGraw-Hill, 1986.", "links": null } }, "ref_entries": { "TABREF0": { "type_str": "table", "content": "
\u4ef6\u4e2d\uff0c\u8fa8\uf9fc\uf961\u4e26\uf967\u9ad8\uff0c\u4e26\uf967\u7b26\u5408\u9ad8\u6e96\u78ba\u6027\u7684\u8981\u6c42[11]\u3002\u800c\u8a9e\u97f3\u6587\u5b57\u6aa2\uf96a\uff0c\u5247\u662f\u5728 \u7279\u5b9a\u8a9e\u97f3\u6a21\u578b\u4e0b\uff0c\u5148\u9032\ufa08\u8a9e\u97f3\u8fa8\uf9fc\uff0c\u518d\u4ee5\u8fa8\uf9fc\u51fa\uf92d\u7684\u6587\u5b57\u9032\ufa08\u6aa2\uf96a[3]\u3002\u7531\u65bc\u76ee \u8a9e\u97f3\u8a0a\u865f \u97f3\u7bc0 \u7d50\u679c \u6211 Syllable \u8655\uf9e4\u3002 (3)\uf974\u4f7f\u7528\u8005\u5c0d\u67e5\u8a62\u51fa\uf92d\u7684\u65b0\u805e\u611f\u8208\u8da3\uff0c\u5247\u9ede\u9078\u8a72\u689d\u65b0\u805e\u5f8c\uff0c\u504f\u4e0b\u65b9\u7684\u767d \u97f3\u7bc0\u8f49\u6587\u5b57 \u2026\u2026\u2026... \u2026\u2026\u2026... (3)\u3010\u904e\uf984\uf967\u9700\u8981\u65b0\u805e\u3011 \u8272\u5340\u584a\u5373\u51fa\u73fe\u5c0d\u61c9\u7684\u65b0\u805e\u5167\u5bb9\u3002
TEL: (03)5715131-3582 \u524d\u6587\u5b57\u6bd4\u5c0d\u6280\u8853\u6210\u719f\uff0c\u6545\u6b64\u6cd5\u7684\u95dc\u9375\u5728\u65bc\u8a9e\u97f3\u8fa8\uf9fc\u7684\u597d\u58de\uff0c\u8fa8\uf9fc\u6587\u5b57\u5167\u5bb9\u5247\u53ef\u5341 \u6458\u8981\uff1a Tree Net \u2026\u2026\u2026... state state state \u2026\u2026\u2026... \u522a\u9664\u6a19\u984c\u6a19\u9ede\u7b26\u865f\uff0c\u524d\u8655\uf9e4\u7cfb\u7d71\u624d\u53ef\u4ee5\u5c0d\u7d14\u4e2d\u6587\u5b57\u9032\ufa08\u6a19\u6ce8 \u5206\u5ee3\u6cdb\u3002 39\u7dad\u7684\u6885\u723e\u5012\u983b\u8b5c\uf96b\uf969 Viterbi Search using \u311b \u975c \u3128 \u975c\u97f3 model model model 1. \u65b0\u805e\u524d\u8655\uf9e4 \u904e\uf984\u6f01\u696d\u6c23\u8c61\u9019\uf9d0\u578b\u7684\u65b0\u805e\uff0c\u5982\u679c\uf967\u9700\u8981\u4e5f\u53ef\u4ee5\uf96d\uf976\u9019\u6b65\u9a5f\u3002 (4)\u4f7f\u7528\u8005\u4e5f\u53ef\u7d93\u7531\u5728\u4e0b\u65b9\u767d\u8272\u5340\u584a\u4e2d\u6309\u4e0b\uf904\u9f20\u5de6\u9375\uf92d\u807d\u53d6\u65b0\u805e\u7684\u5167 1.\u8072\u5b78\u6a21\u578b Hidden Markov Model 2.\u8a9e\u8a00\u6a21\u578b \u2026\u2026\u2026.. . \u2026\u2026\u2026... \u2026\u2026\u2026.. . (4)\u3010\u904e\uf984\u6a19\u984c\u6a19\u9ede\u7b26\u865f\u3011 \u5bb9\uff0c\u8a72\u5167\u5bb9\u662f\u4ee5\u8a9e\u97f3\u5408\u6210\u7684\u65b9\u5f0f\u5373\u6642\u7522\u751f\u7684\u3002
\u5728\u6b64\u5831\u544a\u4e2d\uff0c\u6211\u5011\u5be6\u4f5c\uf9ba\u4e00\u500b\u7d50\u5408\u96b1\u85cf\u5f0f\u99ac\u53ef\u592b\u6a21\u578b(Hidden Markov Model, HMM) \u70ba\u57fa\u790e\u7684 HTK(HMM Toolkit)\u548c\u7db2\u9801\u8cc7\uf9be\u6aa2\uf96a\u6280\u8853\u7684\u7dda\u4e0a\u65b0\u805e\u8a9e\u97f3\u8cc7\uf9be\u6aa2\uf96a\u7cfb \u800c\u6240\u8b02\u7684\u8a9e\u97f3\u8fa8\uf9fc\uff0c\u4e3b\u8981\u662f\u7528\uf92d\u8fa8\uf9fc\u51fa\u8072\u97f3\u7684\u6587\u5b57\u5167\u5bb9\u70ba\u4f55\u3002\u4e00\u822c\uf92d\uf96f\uff0c\u8a9e \u97f3\u8fa8\uf9fc\u7684\u8fa8\uf9fc\u6210\u529f\uf961\u548c\u8fa8\uf9fc\u5167\u5bb9\u7684\u7bc4\u570d\u6709\u5f88\u5927\u95dc\u4fc2\u3002\u5728\u5927\uf9b4\u57df\u7684\u6587\u5b57\u8fa6\uf9fc\u4e2d\uff0c\u8fa8 \u5716\u8868 2 \u8a9e\u97f3\u8fa8\uf9fc\u7684\u6f14\u7b97\u6cd5 \u5176\u4e2d\u6240\u63a1\u7528\u7684\u65b9\u6cd5\u548c\uf9e4\uf941\u5982\u4e0b\uff1a MFCC \u683c\u5f0f MFCC \u683c\u5f0f \u97f3\u3002\u5c07\u904e\uf984\u5b8c\u7684\u6587\u5b57\u5b58\u653e\u65bc\u8cc7\uf9be\u5eab\u7684 news_title_pure \uf91d\u4f4d\u3002 MFCC \u683c\u5f0f \u2026\u2026\u2026.. . \u2026\u2026\u2026.. . \u2026\u2026\u2026... stream stream stream \u2026\u2026\u2026... (5)\u3010\u904e\uf984\u5167\u6587\u6a19\u9ede\u7b26\u865f\u3011
\u7d71\u3002\u4e00\u822c\u7684\u7db2\u9801\u8cc7\uf9be\u6aa2\uf96a(\u5982 google)\u9808\u4f7f\u7528\u8005\u8f38\u5165\u76f8\u95dc\u6587\u5b57\uff0c\u624d\u5f97\u4ee5\u6587\u5b57\u6bd4\u5c0d \u65b9\u5f0f\u9032\ufa08\u6aa2\uf96a\u3002\u5728\u6b64\u6211\u5011\u5247\u5617\u8a66\u52a0\u5165\u8a9e\u97f3\u8fa8\uf9fc\u7684\u6280\u8853\u8b93\u4f7f\u7528\u8005\uf901\uf9e0\u9032\ufa08\u6aa2\uf96a\u3002\u672c \u7cfb\u7d71\u5206\u6210\u65b0\u805e\u524d\u8655\uf9e4\u53ca\u8a9e\u97f3\u67e5\u8a62\uf978\u968e\u6bb5\u3002\u5728\u8fa8\uf9fc\u5167\u5bb9\u56fa\u5b9a\uff0c\u9ad8\u6e96\u78ba\ufa01\u7684\u8fa8\uf9fc\u7d50\u679c \uf9fc\u7d50\u679c\u5f80\u5f80\u51fa\u73fe\u76f8\u8fd1\u4f46\u975e\u6b63\u78ba\u7684\u7b54\u6848\uff0c\u9019\u5728\u76ee\u524d\u4ecd\u662f\u5f88\u96e3\u514b\u670d\u7684\u554f\u984c\u3002 \u9673\u60e0\u82ac \u9673 \u60e0 \u82ac (6)\u3010\uf95a\u51fa\u8cc7\uf9be\u5eab\u3011 \u76ee\u524d\u6b64\uf9b4\u57df\u6700\u6709\u540d\u7684\u6a21\u578b\u70ba HMM\u3002\u85c9\u7531\u7279\u5b9a\u8a9e\u8a00\u8a9e\uf9be\u7684\u8a13\uf996\uff0c\u6211\u5011\u53ef\u4ee5\uf9dd\u7528 \u7528\u9ad8\u65af\u6df7 \u7528\u9ad8\u65af\u6df7 \u7528\u9ad8\u65af\u6df7 \u522a\u9664\u5167\u6587\u6a19\u9ede\u7b26\u865f\uff0c\u524d\u8655\uf9e4\u7cfb\u7d71\u624d\u53ef\u4ee5\u5c0d\u7d14\u4e2d\u6587\u5b57\u9032\ufa08\u6a19\u6ce8 1. \u8a9e\u8a00\u6a21\u578b Tree Net \u628a\u6bcf\u500b\u55ae\u97f3\u7bc0\u8996\u70ba\u4e00\u500b\u7bc0\u9ede\uff0c\u7bc0\u9ede\u548c\u7bc0\u9ede\u9593\u76f8\uf99a\u95dc\u4fc2\u7684\u6a39\uf9fa\u7d50\u69cb\u3002\u5716\uf9b5\u5982\u4e0b\uff1a \u5408\u6a21\u578b\uf92d \u5efa\uf9f7\u6a21\u578b \u5408\u6a21\u578b\uf92d \u5efa\uf9f7\u6a21\u578b \u5408\u6a21\u578b\uf92d \u5efa\uf9f7\u6a21\u578b mixture mixture \u97f3\u3002\u5c07\u904e\uf984\u5b8c\u7684\u6587\u5b57\u5b58\u653e\u65bc\u8cc7\uf9be\u5eab\u7684 news_content_pure \uf91d\u4f4d\u3002 mixture \u65b0\u805e\u6aa2\uf96a\u6309\u9215
\u4e0b\uff0c\u672c\u7cfb\u7d71\u7279\u5225\u9069\u7528\u65bc\u624b\u6a5f\u3001PDA\u3001\u5d4c\u5165\u5f0f\u7cfb\u7d71\u7b49\u5c0f\u578b\u3001\uf967\uf9e0\u4ee5\u624b\u64cd\u4f5c\u8f38\u5165\u7684\u88dd HTK[7](Hidden Markov Model ToolKit)\u5be6\u4f5c\u51fa\u67d0\u4e00\u7279\u5b9a\uf9b4\u57df\u7684\u9ad8\u6e96\u78ba\ufa01\u7684\u8a9e\u97f3 \u9673\u96c5\u79c0 \u96c5 \u79c0 \u5716\u8868 6 Model, State, Stream \u548c Mixture \u793a\u610f\u5716 \u5716\u8868 7 \u4ee5 GMM \u5efa\uf9f7 syllable \u7684 CHMM \uf9ca\u7a0b\u793a\u610f\u5716 \u5728\u9032\ufa08\u6a19\u6ce8\u97f3\u524d\u5c07\u8cc7\uf9be\u5eab\u65b0\u589e\u7684\uf978\uf91d\u4f4d\u5167\u7684\u8cc7\uf9be\u8f49\u6210\u6587\u5b57\u6a94\u3002 Scroll Bar\uff1a\u986f\u793a\uf93f\u97f3\u9032\ufa01
\u7f6e\u3002\u672c\u7cfb\u7d71\u4ea6\u7d93\u6e05\u5927\u76f2\u53cb\u6703\u7684\u76f2\u4eba\u670b\u53cb\u8a66\u7528\uff0c\u53cd\u61c9\u5341\u5206\uf97c\u597d\u3002 \u8fa8\uf9fc\u7cfb\u7d71\uff0c\u6bd4\u65b9\uf96f\u5510\u8a69\u4e09\u767e\u9996\u7684\u8a9e\u97f3\u8fa8\uf9fc\uff0c\u5176\u6e96\u78ba\uf961\u63a5\u8fd1\u4e5d\u6210\u4e5d[10]\u3002 \u9673\u96c5\uf9ad \uf9ad (7)\u3010\u6a19\u6ce8\u97f3\u3011 3. \u8fa8\uf9fc\u65b9\u6cd5 \u67e5\u8a62\u7d50\u679c\u7684\u6a19\u984c\u986f\u793a\u5340
\u95dc\u9375\u8a5e:\u8a9e\u97f3\u8fa8\uf9fc\u3001\u8cc7\uf9be\u6aa2\uf96a\u3001HMM\u3001Viterbi Search\u3001\u65b0\u805e\u6aa2\uf96a\u3002 \u7d9c\u5408\u4ee5\u4e0a\u89c0\u9ede\uff0c\u6211\u5011\u5be6\u4f5c\uf9ba\u4e00\u500b\u7d50\u5408 HTK \u548c\u65b0\u805e\u7db2\u9801\u5167\u5bb9\u7684\u6aa2\uf96a\u6280\u8853\uff0c\u5e0c\u671b \u5716\u8868 3 Tree Net \u6a94\u793a\u610f \u6211\u5011\u6839\u64da Tree Net \u7684\uf937\u5f91\u9032\ufa08\u4ee5 Viterbi Search\uff0c\u4ee5\u8fa8\uf9fc\u51fa\u6a5f\uf961\u6700\u9ad8\u7684\uf937 \u91dd\u5c0d\u5f9e\u8cc7\uf9be\u5eab\u8f49\u51fa\uf92d\u7684\u6587\u5b57\u6a94\u9032\ufa08\u6a19\u6ce8\u97f3\u3002
(8)\u3010\u5efa\uf9f7 NET \u6a94\u6848\u3011 \u5f91\u3002\u5176\u4e2d\u6211\u5011\u4e5f\u52a0\u4e0a\uf9ba Beam Search(\u5149\u675f\u641c\u5c0b\u6cd5)\u7684\u4f5c\u6cd5\u4ee5\u9032\ufa08\u52a0\u901f[2]\u3002\u5149\u901f 1 \u524d\u8a00 \u80fd\u9054\u5230\u4e00\u500b\u4ee5\u8a9e\u97f3\u70ba\u57fa\u790e\u7684 News Google\u3002 2. \u8072\u5b78\u6a21\u578b Hidden Markov Model \u641c\u5c0b\u6cd5\u5728\u641c\u5c0b\u904e\u7a0b\u4e2d\u6703\u6162\u6162\u4e1f\u68c4\u4f4e\u6a5f\uf961\u7684\u641c\u5c0b\u76ee\u6a19\uff0c\u4f7f\u5f97\u6108\u5f8c\u9762\u7684\u6bd4\u5c0d\u901f\ufa01\u6703\u6108 \u91dd\u5c0d\u6a19\u597d\u6ce8\u97f3\u7684\u6a94\u6848\u5efa\uf9f7 Tree Net\uff0c\u4ee5\u4f9b\u8a9e\u97f3\u8fa8\uf9fc\u7a0b\u5f0f\u67e5\u8a62\u3002
\u76ee\u524d\u7684\u7db2\u969b\u7db2\uf937\u4e2d\uff0cwww.google.com[6]\u662f\u6bcf\u500b\u4eba\ufa26\uf967\u53ef\u6216\u7f3a\u7684\u5de5\u5177\uff0c\u5176\u63d0 2 \u8a9e\u97f3\u65b0\u805e\u6aa2\uf96a\uf9e4\uf941\u80cc\u666f \u96b1\u85cf\u5f0f\u99ac\u53ef\u592b\u6a21\u578b\u57fa\u672c\u4e0a\u662f\u4e00\u7a2e\u96d9\u91cd\u96a8\u6a5f\u904e\u7a0b\uff0c\u800c\u4e4b\u6240\u4ee5\u7a31\u70ba\u96b1\u85cf\u5f0f\u662f\u56e0\u70ba \u52a0\u5feb\uff0c\u6b64\u6cd5\u53ef\u6709\u6548\u6e1b\u5c11\u641c\u5c0b\u6642\u9593\u4e14\uf967\u6703\u72a7\u7272\u592a\u591a\u6e96\u78ba\u6027[5]\u3002 (9)\u3010\u767c\u5e03\u5230\u8cc7\uf9be\u5eab\u4f3a\u670d\u5668\u3011 \u5167\u6587\u986f\u793a\u5340
\u4f9b\u7684\u6e96\u78ba\u6027\u548c\u8cc7\uf9be\u7684\u53ef\u7528\u6027\u4e00\u76f4\u70ba\u4eba\u7a31\u9053\uff0c\u582a\u7a31\u70ba\u6587\u5b57\u6aa2\uf96a\u7684\u7ff9\u695a\uff0c\u4e5f\u56e0\u6b64\uff0c\u7db2 \u672c\u7cfb\u7d71\u4f7f\u7528\uf9ba\u8a9e\u97f3\u8fa8\uf9fc\u3001\u6587\u5b57\u6bd4\u5c0d\u548c\u8a9e\u97f3\u5408\u6210\u4e09\u7a2e\u6280\u8853\u3002\uf9ca\u7a0b\u5716\u5982\u4e0b\uff1a \u5176\u4e2d\u6709\u4e00\u7d44\u96a8\u6a5f\u904e\u7a0b\u662f\u96b1\u85cf\u7684\uff0c\u770b\uf967\ufa0a\u7684\uff0c\u5728\u8a9e\u97f3\u4e2d\u5c31\u5982\u540c\u4eba\uf9d0\u5728\u767c\u8072\u7684\u904e \uf901\u65b0\u8cc7\uf9be\u5eab\u4f3a\u670d\u5668\u7684\u8cc7\uf9be\u3002
\u7a0b\u4e2d\u5176\u767c\u8072\u5668\u5b98\uf9fa\u614b\u8b8a\u5316\u662f\u770b\uf967\ufa0a\u7684\uff0c\u597d\u6bd4\u5589\u56a8\u3001\u820c\u982d\u8207\u53e3\u8154\u7684\u8b8a\u5316\u662f\uf967\u53ef 2.2 \u8a9e\u97f3\u5408\u6210 (10)\u3010\u767c\u5e03\u8cc7\uf9be\u5230\u8fa8\uf9fc\u7cfb\u7d71\u3011 \u5716\u8868 12 \u65b0\u805e\u8a9e\u97f3\u67e5\u8a62\u4ecb\u9762 \u969b\u7db2\uf937\u4e0a\u7684\u8cc7\u8a0a\u6210\uf9ba\u4e00\u500b\u7121\u6240\uf967\u5305\u7684\u8cc7\uf9be\u5eab\u3002\u800c\u5728\u6b64\u540c\u6642\uff0c\u76f8\u95dc\u7684\u591a\u5a92\u9ad4\u6aa2\uf96a\u6280 \u8853\u4e5f\u76f8\u7e7c\u767c\u8868[1]\uff0c\u986f\u793a\uf9ba\u591a\u5a92\u9ad4\u65b9\u9762\u7684\u6aa2\uf96a\u9700\u6c42\u3002\u800c\u7531\u65bc\u8a9e\u97f3\u7684\uf965\uf9dd\u6027\u548c\u53ef\u7528 \u6587\u5b57\u6aa2\uf96a \u67e5\u8a62\u4ecb\u9762 \u4f7f\u7528\u8005 TTS \u8a9e\u97f3\u5408\u6210 \u8072\u97f3\u8a0a\u865f \u8a9e\u97f3\u8fa8\uf9fc \u80fd\u5f9e\u53ef\u89c0\u6e2c\u7684\u8a9e\u97f3\u8a0a\u865f\u5e8f\uf99c\u770b\u51fa\uf92d\u7684\u3002\u800c\u53e6\u4e00\u7d44\u96a8\u6a5f\u904e\u7a0b\u7a31\u70ba\u89c0\u6e2c\u5e8f\uf99c (observation sequence) \uff0c \u5b83 \u662f \u7531 \uf9fa \u614b \u89c0 \u6e2c \u6a5f \uf961 (state observation \u5728\u8f38\u51fa\u65b9\u9762\u6211\u5011\u4f7f\u7528\u548c\u9ec3\u7d39\u83ef\uf934\u5e2b\u5408\u4f5c\u7684\u8a9e\u97f3\u5408\u6210\u6280\u8853\u3002\u6b64\u5408\u6210\u65b9\u5f0f\u662f\uf99a\u63a5\u5f0f\u7684 \u5982\u5716\u8868 10 \u6240\u793a\uff0c\u672c\u7a0b\u5f0f\u5206\u70ba\u5341\u5927\u6b65\u9a5f.\u3002 \u3010\u65b0\u805e\u524d\u8655\uf9e4\u3011\u6309\u9215\u5247\u662f 2. \u8a9e\u97f3\u67e5\u8a62\u65b0\u805e \u5408\u6210\u70ba\u57fa\u790e(Concatenation-Based)\uff0c\u57fa\u672c\uf9ca\u7a0b\u5982\u4e0b\uff1a \u5716\u8868 9 \u65b0\u805e\u524d\u8655\uf9e4\u4ecb\u9762 \uf901\u65b0\u8fa8\uf9fc\u7cfb\u7d71\u7684\u8fa8\uf9fc\u6838\u5fc3\u3002 4 \u7d50\uf941
\u6027(\u76f8\u5c0d\u65bc\u4ee5\u5167\u5bb9\u70ba\u4e3b\u7684\u5f71\u50cf\u6aa2\uf96a)\uff0c\u8a9e\u97f3\u65b9\u9762\u7684\u6aa2\uf96a\u65b9\u6cd5\u5df2\u6210\uf9ba\u591a\u5a92\u8cc7\u8a0a\u6aa2\uf96a\u7684 \u91cd\u8981\u7814\u7a76\u3002 \u8a9e \u97f3 \u6aa2 \uf96a \u7684 \u65b9 \u6cd5 \u53ef \u5206 \u70ba \uf978 \u7a2e \uff0c \u4e00 \u70ba \u8a9e \u97f3 \u6587 \u4ef6 \u6aa2 \uf96a (Spoken Document Retrieval)\uff0c\u4e00\u70ba\u8a9e\u97f3\u6587\u5b57\u6aa2\uf96a(Speech Recognition and Retrieval)\u3002\u524d\u8005\uf967 probability)\uf92d\u63cf\u8ff0\u5728\u6bcf\u500b\uf9fa\u614b\u4e0b\u89c0\u6e2c\u5230\u5404\u7a2e\u8a9e\u97f3\u7279\u5fb5\uf96b\uf969\u7684\u6a5f\uf961\u5206\u4f48\u3002 \u6587\u5b57\u683c\u5f0f\u8f38\u5165 \u6309\u4e0b\u5f8c\u5373\u53ef\u57f7\ufa08 Step0 ~ Step9\u3002\u4ee5\u4e0b\u4f9d\u5e8f\u4ecb\u7d39\u5404\u529f\u80fd\uff1a \u5728\u672c\u7bc7\u5831\u544a\u4e2d\uff0c\u6211\u5011\u4ecb\u7d39\uf9ba\u4e00\u500b\u300c\u7dda\u4e0a\u65b0\u805e\u8a9e\u97f3\u8cc7\uf9be\u6aa2\uf96a\u7cfb\u7d71\u300d \u3002\u6b78\u7d0d\u7d50\u679c\uff0c \u6211\u5011\u4ee5 Borland C++ Builder 5.0 \u5efa\u69cb\u65b0\u805e\u8a9e\u97f3\u67e5\u8a62\u4ecb\u9762\uff0c\u5982\u5716\u8868 \u5716\u8868 1 \u8a9e\u97f3\u65b0\u805e\u6aa2\uf96a\u670d\u52d9\uf9ca\u7a0b HMM \u7684\uf9fa\u614b\u89c0\u6e2c\u6a5f\uf961\u51fd\u5f0f ) ( t j o b \u662f\u63a1\u7528\u9ad8\u65af\u6df7\u5408\u5bc6\ufa01\u51fd\uf969\u6216\u7a31\u9ad8\u65af\u6df7\u5408\u6a21\u578b (1)\u3010\u65b0\u805e\u6a94\u6848\u4e0b\u8f09\u3011 \u5728\u6b64\uf99c\u51fa\u6b64\u7cfb\u7d71\u7684\u7279\u6027\uff1a 11\u3002\u6b64\u4ecb\u9762\u5206\u6210\u6a19\u984c\u67e5\u8a62\u53ca\u5167\u6587\u67e5\u8a62\uf978\u90e8\u4efd\uff0c\u9867\u540d\u601d\u7fa9\uff0c\u6a19\u984c\u67e5\u8a62\u70ba\u627e \u6587\u5b57\u6bd4\u5c0d\u90e8\u4efd\uff0c\u7531\u6211\u5011\u53ea\u5c0d\u65b0\u805e\u6a19\u984c\u90e8\u4efd\u4f5c\u6bd4\u5c0d\uff0c\u56e0\u6b64\u4ee5\u4e0b\u4ee5\u8a9e\u97f3\u8fa8\uf9fc\u548c\u5408 \u6210\u4f5c\u70ba\u4ecb\u7d39\u91cd\u9ede\u3002 2.1 \u8a9e\u97f3\u8fa8\uf9fc\u90e8\u4efd (Gaussian Mixture Model, GMM)\uf92d\u8a08\u7b97\uf99a\u7e8c\u6a5f\uf961\u5bc6\ufa01\uff0c\u56e0\u6b64\u6bcf\u4e00\u500b\u8072\u97f3\u55ae \u5143(Model)\u7686\u6709\u4e00\u7d44 Continuous HMM \uf96b\uf969\u3002 Wave Maker \u805e\u3002 (1)\u6309\u4e0b\u4e00\u500b\u6aa2\uf96a\u6309\u9215\uff0c\u7cfb\u7d71\u6703\u4ee5\u8a9e\u97f3\u7684\u65b9\u5f0f\u63d0\u793a\u4f7f\u7528\u8005\u6e96\u5099\uf93f\u97f3\uff0c\uf93f\u97f3 \u5716\u8868 4 \u70ba Model, State, Stream \u548c Mixture \u7684\u968e\u5c64\u793a\u610f\u5716\uff0c\u5716\u8868 5 \u5247 Speech Synthesis \u8a5e\u5f59\u8cc7\uf9be\u5eab \u4ee5 PERL \u7a0b\u5f0f\uff0c\u5f9e\u7db2\uf937\u4e0a\u6293\u53d6\u7576\u65e5\u7684\u65b0\u805e\uff0c\u76ee\u524d\u7cfb\u7d71\u9810\u8a2d\u503c\u70ba\u6293 \u7b26\u5408\u95dc\u9375\u5b57\u7684\u6a19\u984c\uff0c\u800c\u5167\u6587\u67e5\u8a62\u5247\u662f\u53ea\u8981\u5167\u6587\u6709\uf906\u5b50\u7b26\u5408\u95dc\u9375\u5b57\u5373\u6703\u986f 1. \u8a9e\u97f3\u8f38\u5165\uff1a\uf967\u9375\u76e4\u7b49\u9808\u5176\u4ed6\u5de5\u5177\uff0c\u5373\u53ef\u5c07\u67e5\u8a62\u5167\u5bb9\u8f38\u5165\u3002 \u8072\u97f3\u6ce2\u5f62 \u8f49\u63db\u6587\u5b57\u6a94\u6848\u6210\u534a\u6210\u54c1\u7684 Raw Data \u53d6\u4e2d\u570b\u6642\u5831\u3001\u53f0\u7063\u65b0\u751f\u5831\u3001\u4e2d\u592e\u793e\u65b0\u805e\u3001\u65b0\uf92a\u7db2\u65b0\u805e\u7b49\u56db\u5bb6\u7db2\u7ad9\u7684\u65b0 \u793a\u51fa\uf92d\u3002\u4ee5\u4e0b\u4ecb\u7d39\u64cd\u4f5c\u6642\u5927\uf976\u7684\uf9ca\u7a0b\uff1a 2. \u5feb\u901f\u6aa2\uf96a\uff1a\u85c9\u7531 offline \u7684\u6a19\u984c\uf96a\u5f15\u548c\u5373\u6642\u7684\u8a9e\u97f3\u8fa8\uf9fc\u3001\u6587\u5b57\u6bd4\u5c0d\uff0c\u63d0\u4f9b\u65b0
\u8003\u616e\u5230\u8a9e\u97f3\u6a21\u578b\uff0c\u76f4\u63a5\u4ee5\u8a9e\u97f3\u7684\u7279\u5fb5\uf96b\uf969\uff0c\u5728\u53e6\u4e00\u8a9e\u97f3\u6587\u4ef6\u4e2d\u9032\ufa08\u6bd4\u5c0d\uff0c\u5e0c\u671b\u627e \u4e00\u822c\u800c\u8a00\uff0c\u8a9e\u97f3\u8fa8\uf9fc\u7684\u6f14\u7b97\u6cd5\uf9ca\u7a0b\u5982\u4e0b\u5716\u6240\u793a\uff1a \u4ee5\uff02\u6211\uff02\u6b64\u4e00 syllable \u70ba\uf9b5\uff0c\u793a\u7bc4 CHMM \u7684\u5efa\uf9f7\u65b9\u5f0f\u3002 \u5c07\u6240\u5f97\u7684 Raw Data \u518d\u8f49\u63db\u6210\u7b26\u5408 Wave \u683c\u5f0f\u7684\u6a94\u6848 (2)\u3010\u65b0\u589e\u8cc7\uf9be\u5eab\uf91d\u4f4d\u3011 \u6642\u9593\u70ba\u4e09\u79d2\u9418\u3002 \u805e\u6a19\u984c\u7684\u5feb\u901f\u6aa2\uf96a\u3002
\u51fa\u6700\u63a5\u8fd1\u7684\u8a9e\u97f3\u5167\u5bb9\u3002\u9019\u6a23\u7684\u6aa2\uf96a\u65b9\u5f0f\u96d6\u53ef\u8de8\u8d8a\u8a9e\u8a00\u6a21\u578b\uff0c\u4f46\u5728\u9577\u6642\u9593\u7684\u8a9e\u97f3\u6587 \u8f38\u51fa Wave \u6a94\u6848 \u505a\u5b8c\u4e0a\u4e00\u6b65\u9a5f\u6293\u53d6\u7576\u65e5\u65b0\u805e\u5b8c\u6210\u5f8c\uff0c\u5728 Access \u8cc7\uf9be\u5eab\u4e2d\u65b0\u589e\uf978 (2)\uf93f\u5b8c\u97f3\u5f8c\u8fa8\uf9fc\u7cfb\u7d71\u5247\u958b\u59cb\u8fa8\uf9fc\u8a9e\u97f3\uff0c\u800c\u5f8c\u5c07\u7d50\u679c\u986f\u793a\u5728\u504f\u4e0a\u65b9\u7684\u767d\u8272 3. \u8a9e\u97f3\u8f38\u51fa\uff1a\u4f7f\u7528 Text-To-Speech \u7684\u8a9e\u97f3\u5408\u6210\uff0c\u5c07\u67e5\u8a62\u6240\u5f97\u65b0\u805e\u9032\ufa08\u64ad\u5831\u3002
\u5716\u8868 8 \u8a9e\u97f3\u5408\u6210\uf9ca\u7a0b \u500b\uf91d\u4f4d\uff0c\u5373 news_title_pure \u53ca news_content_pure\uff0c\u4ee5\uf965\u63a5\u4e0b\uf92d\u7684 \u5340\u584a\u5167\u3002 4. \u5b9a\u6642\uf901\u65b0\uff1a\u6bcf\u65e5\u56fa\u5b9a\u6642\u9593\uf901\u65b0\u7db2\u9801\u4e0a\u5373\u6642\u65b0\u805e\u3002
", "text": "\u9673\u6c5f\u6751 \uf90f\u745e\uf9f3 \u5f35\u667a\u661f \u570b\uf9f7\u6e05\u83ef\u5927\u5b78 \u8cc7\u8a0a\u5de5\u7a0b\u7cfb \u65b0\u7af9\u5e02\u5149\uf966\uf937\u4e8c\u6bb5 101 \u865f E-mail : {jtchen,roro,jang}@wayne.cs.nthu.edu.tw", "num": null, "html": null } } } }