{
    "paper_id": "O14-1007",
    "header": {
        "generated_with": "S2ORC 1.0.0",
        "date_generated": "2023-01-19T08:04:45.501392Z"
    },
    "title": "\u57fa\u65bc\u767c\u97f3\u77e5\u8b58\u4ee5\u5efa\u69cb\u983b\u8b5c HMM \u4e4b\u570b\u8a9e\u8a9e\u97f3\u5408\u6210\u65b9\u6cd5 A Mandarin Speech Synthesis Method Using Articulation-knowledge Based Spectral HMM Structure",
    "authors": [
        {
            "first": "",
            "middle": [],
            "last": "\u53e4\u9d3b\u708e",
            "suffix": "",
            "affiliation": {
                "laboratory": "",
                "institution": "National Taiwan University of Science and Technology",
                "location": {}
            },
            "email": ""
        },
        {
            "first": "\u3001\u8cf4\u540d\u5f65",
            "middle": [
                "*"
            ],
            "last": "\u3001\u6d2a\u5c09\u7fd4",
            "suffix": "",
            "affiliation": {
                "laboratory": "",
                "institution": "National Taiwan University of Science and Technology",
                "location": {}
            },
            "email": ""
        },
        {
            "first": "",
            "middle": [],
            "last": "\u3001\u9673\u5f65\u6a3a",
            "suffix": "",
            "affiliation": {
                "laboratory": "",
                "institution": "National Taiwan University of Science and Technology",
                "location": {}
            },
            "email": ""
        },
        {
            "first": "Hung-Yan",
            "middle": [],
            "last": "Gu",
            "suffix": "",
            "affiliation": {
                "laboratory": "",
                "institution": "National Taiwan University of Science and Technology",
                "location": {}
            },
            "email": ""
        },
        {
            "first": "Ming-Yen",
            "middle": [],
            "last": "Lai",
            "suffix": "",
            "affiliation": {
                "laboratory": "",
                "institution": "National Taiwan University of Science and Technology",
                "location": {}
            },
            "email": ""
        },
        {
            "first": "Wei-Siang",
            "middle": [],
            "last": "Hong",
            "suffix": "",
            "affiliation": {
                "laboratory": "",
                "institution": "National Taiwan University of Science and Technology",
                "location": {}
            },
            "email": ""
        },
        {
            "first": "Yan-Hua",
            "middle": [],
            "last": "Chen",
            "suffix": "",
            "affiliation": {
                "laboratory": "",
                "institution": "National Taiwan University of Science and Technology",
                "location": {}
            },
            "email": ""
        }
    ],
    "year": "",
    "venue": null,
    "identifiers": {},
    "abstract": "In this paper, a new HMM structure is proposed to work with a limited training corpus in order to obtain improved synthetic-speech fluency. Spectral fluency is improved because this HMM structure can model the context-dependent spectral characteristics of a speech unit. In addition, instead of using a decision tree to cluster contexts, the knowledge of phoneme articulation is based to cluster contexts and reduce the enormous quantity of context combinations. To evaluate the proposed HMM structure, we construct three Mandarin speech synthesis systems each uses one different HMM structure for comparisons. In these systems, the prosodic parameters are all generated with same ANN modules studied previously",
    "pdf_parse": {
        "paper_id": "O14-1007",
        "_pdf_hash": "",
        "abstract": [
            {
                "text": "In this paper, a new HMM structure is proposed to work with a limited training corpus in order to obtain improved synthetic-speech fluency. Spectral fluency is improved because this HMM structure can model the context-dependent spectral characteristics of a speech unit. In addition, instead of using a decision tree to cluster contexts, the knowledge of phoneme articulation is based to cluster contexts and reduce the enormous quantity of context combinations. To evaluate the proposed HMM structure, we construct three Mandarin speech synthesis systems each uses one different HMM structure for comparisons. In these systems, the prosodic parameters are all generated with same ANN modules studied previously",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Abstract",
                "sec_num": null
            }
        ],
        "body_text": [
            {
                "text": "\u8fd1\u5e74\u4f86\u8a31\u591a\u7814\u7a76\u8005\u65e9\u5df2\u5229\u7528\u96b1\u85cf\u5f0f\u99ac\u53ef\u592b\u6a21\u578b(hidden Markov model, HMM)\uff0c\u4f86\u5efa\u9020\u8a9e\u97f3 \u55ae\u5143(\u5982\u97f3\u7d20\u3001\u97f3\u7bc0\u7b49)\u4e4b\u983b\u8b5c\u6f14\u9032(spectrum progression)\u6a21\u578b (Yoshimura et al., 1999; Zen et al., 2007; Yan et al., 2009; Gu et al., 2010) \uff0c\u4e4b\u5f8c\u5728\u5408\u6210\u4e00\u500b\u8a9e\u53e5\u6642\uff0c\u5c31\u6703\u4f7f\u7528\u8a13\u7df4\u5f97\u5230\u7684 HMM \u6a21\u578b\u4f86\u7522\u751f\u4e00\u5e8f\u5217\u7684\u983b\u8b5c\u7279\u5fb5\u5411\u91cf\uff0c\u7136\u5f8c\u4f7f\u7528\u6240\u7522\u751f\u7684\u983b\u8b5c\u7279\u5fb5\u5411\u91cf\u5e8f\u5217\u53bb\u5408\u6210 \u51fa \u8a9e \u97f3 \u4fe1 \u865f \u3002 \u4f7f \u7528 HMM \u4f86 \u4f5c \u8a9e \u97f3 \u4fe1 \u865f \u7684 \u5408 \u6210 \uff0c \u901a \u5e38 \u80fd \u5920 \u7372 \u5f97 \u589e \u9032 \u7684 \u53ef \u7406 \u89e3 \u6027 (intelligibility)\u8207\u6d41\u66a2\u6027(fluency)\u3002\u66f4\u597d\u7684\u662f\uff0cTokuda \u7b49\u4eba\u57fa\u65bc HTK (HMM tool kits)\u6240\u767c \u5c55\u7684 HTS \u8a9e\u97f3\u5408\u6210\u8edf\u9ad4 (Zen et al., 2007) \uff0c\u63d0\u4f9b\u516c\u958b\u7684\u7a0b\u5f0f\u539f\u59cb\u78bc\u3001\u4e26\u4e14\u53ef\u4f9b\u4e0b\u8f09\uff0c\u6240\u4ee5 \u7814\u7a76\u8a9e\u97f3\u5408\u6210\u6642\uff0c\u4f7f\u7528 HTS \u80fd\u5920\u6e1b\u5c11\u8a31\u591a\u6642\u9593\u8207\u6c23\u529b\u3002\u4e0d\u904e\uff0c\u7576\u672a\u4f7f\u7528\u5168\u57df\u8b8a\u7570\u6578(global variance, GV)\u5339\u914d\u6642\uff0cHTS \u8edf\u9ad4\u6240\u7522\u751f\u7684\u983b\u8b5c\u5305\u7d61\u6703\u767c\u751f\u904e\u65bc\u5e73\u6ed1\u7684\u73fe\u8c61\uff0c\u4f7f\u5f97\u5408\u6210\u51fa \u7684\u8a9e\u97f3\u8b8a\u5f97\u60b6\u60b6\u7684(muffled) (Toda & Tokuda, 2005 )\u3002 \u5728\u672c\u8ad6\u6587\u4e2d\uff0c\u6211\u5011\u4e26\u4e0d\u6253\u7b97\u6cbf\u7528\u3001\u4fee\u6539 HTS \u7684\u7a0b\u5f0f\u78bc\uff0c\u56e0\u6b64\u5fc5\u9808\u81ea\u884c\u767c\u5c55 HMM \u6a21 \u578b\u8a13\u7df4\u7684\u7a0b\u5f0f\u3001\u53ca\u983b\u8b5c\u7279\u5fb5\u5411\u91cf\u5e8f\u5217\u4e4b\u7522\u751f\u7a0b\u5f0f\uff0c\u9019\u6a23\u7684\u6c7a\u5b9a\u662f\u56e0\u70ba\uff0c\u6211\u5011\u60f3\u8981\u7814\u767c\u4e00 \u500b\u5177\u6709\u5f48\u6027(flexibility)\u7684\u8a9e\u97f3\u5408\u6210\u7cfb\u7d71\uff0c\u80fd\u5920\u5bb9\u6613\u5730\u64f4\u589e\u984d\u5916\u7684\u529f\u80fd\u3002\u4f8b\u5982\u97f3\u8272\u8f49\u63db (timbre transformation)\u4e4b\u529f\u80fd\uff0c\u80fd\u5920\u628a\u5408\u6210\u8a9e\u97f3\u7684\u97f3\u8272\u5f9e\u6210\u5e74\u5973\u6027\u8f49\u8b8a\u6210\u7537\u5b69 (Gu & Tsai, 2013 )\uff1b\u53e6\u4e00\u9805\u9810\u8a08\u64f4\u589e\u7684\u529f\u80fd\u5247\u662f\uff0c\u540c\u6b65\u5730\u64ad\u653e\u5408\u6210\u51fa\u7684\u8a9e\u97f3\u4fe1\u865f\u53ca\u5176\u5c0d\u61c9\u7684\u62fc\u97f3\u7b26 \u865f\uff0c\u6b64\u529f\u80fd\u53ef\u61c9\u7528\u65bc\u4eba\u5f62\u6a5f\u5668\u4eba\u4e0a\uff0c\u8b93\u8a9e\u97f3\u767c\u8072\u548c\u5634\u578b\u540c\u6b65\u3002\u9664\u4e86\u64f4\u589e\u984d\u5916\u529f\u80fd\u7684\u539f\u56e0 \u4e4b\u5916\uff0c\u6211\u5011\u8207\u5176\u4ed6\u7814\u7a76\u8005\u4f7f\u7528 HTS \u7684\u7d93\u9a57 (Hsia et al., 2010) \uff0c\u767c\u73fe HTS \u6240\u7522\u751f\u7684\u570b\u8a9e\u8a9e \u97f3\u7684\u57fa\u9031\u8ecc\u8de1(pitch contours)\u807d\u89ba\u4e0a\u4e26\u4e0d\u4ee4\u4eba\u6eff\u610f\uff0c\u56e0\u6b64\u672c\u8ad6\u6587\u5efa\u9020\u7684\u570b\u8a9e\u8a9e\u97f3\u5408\u6210\u7cfb \u7d71\uff0c\u5c31\u6c7a\u5b9a\u4f7f\u7528\u4e0d\u540c\u7684\u65b9\u6cd5\u4f86\u7522\u751f\u5404\u500b\u97f3\u7bc0\u7684\u57fa\u9031\u8ecc\u8de1\u3002 \u5728\u5148\u524d\u7684\u4e00\u6b21\u7814\u7a76\u4e2d (Gu et al., 2010) \uff0c\u6211\u5011\u66fe\u5617\u8a66\u53bb\u5efa\u7acb\u97f3\u7bc0\u55ae\u4f4d\u7684 HMM \u6a21\u578b\uff0c \u4ee5\u638c\u63e1\u97f3\u7bc0\u5167\u7684\u983b\u8b5c\u6f14\u9032(spectrum progression)\u65b9\u5f0f\uff0c\u4f46\u662f\u6b64 HMM \u6a21\u578b\u6240\u5408\u6210\u51fa\u7684\u8a9e\u97f3 \u4fe1\u865f\u4e26\u4e0d\u5920\u6d41\u66a2\uff0c\u5728\u97f3\u7bc0\u908a\u754c\u8655\u7684\u983b\u8b5c\u4e0d\u9023\u7e8c(spectral discontinuities)\u60c5\u5f62\u7d93\u5e38\u6703\u88ab\u807d\u51fa \u4f86\uff0c\u6211\u5011\u8a8d\u70ba\u767c\u751f\u983b\u8b5c\u4e0d\u9023\u7e8c\u7684\u539f\u56e0\u662f\uff0c\u8a2d\u5b9a\u7684\u8a9e\u97f3\u55ae\u4f4d\"\u97f3\u7bc0\"\u592a\u5927\uff0c\u4f7f\u5f97\u4e00\u500b\u97f3\u7bc0 X \u548c\u524d\u5f8c\u97f3\u7bc0\u6240\u7d44\u5408\u51fa\u7684\u4e0d\u540c\u6587\u8108\u6578\u91cf\u975e\u5e38\u9f90\u5927\uff0c\u4ee5\u81f3\u65bc\u7121\u6cd5\u70ba\u8a72\u97f3\u7bc0 X \u5efa\u7acb\u826f\u597d\u7684\u6587\u8108 \u76f8\u4f9d HMM \u6a21\u578b\u3002\u56e0\u6b64\uff0c\u672c\u8ad6\u6587\u6c7a\u5b9a\u4f7f\u7528\u8f03\u5c0f\u7684\u8a9e\u97f3\u55ae\u4f4d\uff0c\u5373\u8072\u6bcd\u548c\u97fb\u6bcd\uff0c\u63a5\u8457\u6839\u64da\u6a19 \u8a18\u6a94\u88e1\u8a18\u8f09\u7684\u62fc\u97f3\u7b26\u865f\u5e8f\u5217\uff0c\u53bb\u5efa\u69cb\u6587\u8108\u76f8\u4f9d\u7684 HMM \u6a21\u578b\uff0c\u4e26\u4e14\u4f9d\u64da\u767c\u97f3\u77e5\u8b58\u4f86\u5c0d HMM \u6a21\u578b\u4f5c\u5206\u7d44\u8a13\u7df4\u3002HMM \u7d50\u69cb\u7684\u8a2d\u8a08\u65b9\u5f0f\u5c07\u6703\u5728\u7b2c\u4e8c\u7bc0\u8a73\u7d30\u8aaa\u660e\u3002 \u6211\u5011\u5efa\u9020\u7684\u570b\u8a9e\u8a9e\u97f3\u5408\u6210\u7cfb\u7d71\uff0c\u5176\u5408\u6210\u968e\u6bb5\u4e4b\u8655\u7406\u6d41\u7a0b\u5982\u5716 1 \u6240\u756b\uff0c\u5340\u584a(a)\u9032\u884c\u8f38 \u5165\u6587\u53e5\u4e4b\u5206\u6790\uff1b\u5340\u584a(b)\u4f7f\u7528\u985e\u795e\u7d93\u7db2\u8def(Artificial Neural Network, ANN)\u6a21\u7d44\u4f86\u7522\u751f\u5404\u500b \u97f3\u7bc0\u7684\u57fa\u9031\u8ecc\u8de1\u53c3\u6578\u548c\u6642\u9577(duration)\u503c\uff0c\u8a73\u7d30\u4f5c\u6cd5\u53ef\u53c3\u8003 (Gu & Wu, 2009 )\uff1b\u5340\u584a(c)\u4f9d\u64da \u7b2c\u4e8c\u7bc0\u4ecb\u7d39\u7684\u6311\u9078\u65b9\u6cd5\u4f86\u70ba\u5404\u500b\u8072\u97fb\u6bcd\u9078\u51fa\u5c0d\u61c9\u7684 HMM \u6a21\u578b\uff1b\u5340\u584a(d)\u63a1\u7528 Tokuda \u7b49 \u4eba\u63d0\u51fa\u7684\u65b9\u6cd5 (Yoshimura et al., 1999) \uff0c\u53bb\u8a08\u7b97 HMM \u5404\u72c0\u614b\u5206\u914d\u5230\u7684\u6642\u9577\u97f3\u6846\u6578\uff1b\u5340\u584a (e)\u63a1\u7528\u5148\u524d\u7814\u7a76\u63d0\u51fa\u7684\u52a0\u6b0a\u7dda\u6027\u5167\u63d2\u6cd5 (Gu et al., 2010) ",
                "cite_spans": [
                    {
                        "start": 94,
                        "end": 118,
                        "text": "(Yoshimura et al., 1999;",
                        "ref_id": "BIBREF10"
                    },
                    {
                        "start": 119,
                        "end": 136,
                        "text": "Zen et al., 2007;",
                        "ref_id": "BIBREF11"
                    },
                    {
                        "start": 137,
                        "end": 154,
                        "text": "Yan et al., 2009;",
                        "ref_id": "BIBREF9"
                    },
                    {
                        "start": 155,
                        "end": 171,
                        "text": "Gu et al., 2010)",
                        "ref_id": "BIBREF3"
                    },
                    {
                        "start": 384,
                        "end": 402,
                        "text": "(Zen et al., 2007)",
                        "ref_id": "BIBREF11"
                    },
                    {
                        "start": 538,
                        "end": 558,
                        "text": "(Toda & Tokuda, 2005",
                        "ref_id": "BIBREF7"
                    },
                    {
                        "start": 737,
                        "end": 753,
                        "text": "(Gu & Tsai, 2013",
                        "ref_id": "BIBREF4"
                    },
                    {
                        "start": 852,
                        "end": 871,
                        "text": "(Hsia et al., 2010)",
                        "ref_id": "BIBREF5"
                    },
                    {
                        "start": 972,
                        "end": 989,
                        "text": "(Gu et al., 2010)",
                        "ref_id": "BIBREF3"
                    },
                    {
                        "start": 1452,
                        "end": 1466,
                        "text": "(Gu & Wu, 2009",
                        "ref_id": "BIBREF2"
                    },
                    {
                        "start": 1531,
                        "end": 1555,
                        "text": "(Yoshimura et al., 1999)",
                        "ref_id": "BIBREF10"
                    },
                    {
                        "start": 1601,
                        "end": 1618,
                        "text": "(Gu et al., 2010)",
                        "ref_id": "BIBREF3"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "\u7dd2\u8ad6",
                "sec_num": "1."
            },
            {
                "text": "\u985e\u4f3c\u65bc\u524d\u4e00\u6bb5\u7684\u6558\u8ff0\uff0c\u5982\u679c\u76ee\u524d\u97f3\u7bc0\u6c92\u6709\u8072\u6bcd\u6642\uff0c\u5247\u5728\u6b64\u97f3\u7bc0\u97fb\u6bcd\u7684\u5de6\u908a\uff0c\u5c07\u6703\u9047 \u5230\u524d\u4e00\u97f3\u7bc0\u7684\u97fb\u6bcd\u6216\u975c\u97f3\uff0c\u5728\u6b64\u4e5f\u628a\u524d\u4e00\u97f3\u7bc0\u97fb\u6bcd\u5c3e\u7aef\u4e4b\u53ef\u80fd\u767c\u97f3\u53e3\u5f62\u5206\u6210 11 \u985e\uff0c\u5982\u8868 1 \u6240\u793a\u3002\u53e6\u4e00\u7a2e\u60c5\u6cc1\uff0c\u7576\u97fb\u6bcd\u5de6\u908a\u9047\u5230\u7684\u662f\u672c\u97f3\u7bc0\u7684\u8072\u6bcd\u6642\uff0c\u5728\u6b64\u6211\u5011\u6839\u64da\u8072\u6bcd\u7684\u767c\u97f3 \u4f4d\u7f6e\u628a\u53ef\u80fd\u9047\u5230\u7684\u8072\u6bcd\u5206\u6210 6 \u985e\uff0c\u8a73\u7d30\u5206\u985e\u65b9\u5f0f\u5982\u8868 3 \u6240\u5217\u3002\u8209\u4f8b\u4f86\u8aaa\uff0c/b/, /p/, /m/, /f/ \u7684\u767c\u97f3\u4f4d\u7f6e\u7686\u5728\u5634\u5507\uff0c\u6240\u4ee5\u5b83\u5011\u90fd\u88ab\u5206\u985e\u81f3\"b\"\u985e\u5225\uff1b\u540c\u6a23\u9053\u7406\uff0c/d/, /t/, /n/, /l/\u7684\u767c\u97f3\u4f4d \u7f6e\u7686\u5728\u9f52\u69fd\uff0c\u6240\u4ee5\u628a\u5b83\u5011\u90fd\u5206\u985e\u81f3\"d\"\u985e\u5225\u3002 \u8868 3. \u8072\u6bcd\u4f9d\u767c\u97f3\u4f4d\u7f6e\u4e4b\u5206\u985e Index 0 1 2 3 4 5 Consonant Classes b \u3105 d \u3109 z \u3117 zh \u3113 j \u3110 g \u310d \u8003\u616e\u76ee\u524d\u97f3\u7bc0\u97fb\u6bcd\u7684\u53f3\u908a\u53ef\u80fd\u9047\u5230\u7684\u8a9e\u97f3\u55ae\u5143\uff0c\u7b2c\u4e00\u7a2e\u60c5\u6cc1\u662f\uff0c\u5f8c\u63a5\u7684\u97f3\u7bc0\u6c92\u6709\u8072 \u6bcd\uff0c\u5728\u6b64\u6211\u5011\u628a\u5f8c\u63a5\u97f3\u7bc0\u7684\u97fb\u6bcd\u958b\u982d\u4e4b\u53ef\u80fd\u767c\u97f3\u53e3\u5f62\u5206\u6210 9 \u985e\uff0c\u8a73\u7d30\u5206\u985e\u65b9\u5f0f\u5982\u8868 1 \u4e2d \u7684 9 \u500b\u985e\u5225\uff0c\u4f46\u4e0d\u5305\u542b/n/\u548c/ng/\uff1b\u7b2c\u4e8c\u7a2e\u60c5\u6cc1\u662f\uff0c\u5f8c\u63a5\u7684\u97f3\u7bc0\u5177\u6709\u8072\u6bcd\uff0c\u5728\u6b64\u6211\u5011\u628a\u5f8c \u63a5\u97f3\u7bc0\u8072\u6bcd\u7684\u53ef\u80fd\u767c\u97f3\u53e3\u5f62\u5206\u6210 6 \u985e\uff0c\u8a73\u7d30\u5206\u985e\u65b9\u5f0f\u5982\u8868 3 \u6240\u5217\uff1b\u7b2c\u4e09\u7a2e\u60c5\u6cc1\u662f\uff0c\u5f8c\u63a5 \u97f3\u7bc0\u4e0d\u5b58\u5728(\u5373\u76ee\u524d\u97f3\u7bc0\u662f\u8a9e\u53e5\u7684\u6700\u5f8c\u97f3\u7bc0)\uff0c\u76f8\u7576\u65bc\u5f8c\u9762\u929c\u63a5\u7684\u662f\u975c\u97f3\u3002\u7576\u628a\u97fb\u6bcd\u524d\u5f8c \u53ef\u80fd\u9047\u5230\u7684\u8072\u3001\u97fb\u6bcd\u767c\u97f3\u53e3\u5f62\u4f5c\u4e86\u5206\u985e\u4e4b\u5f8c\uff0c\u518d\u8003\u616e\u570b\u8a9e\u5171\u6709 37 \u500b\u97fb\u6bcd\uff0c\u7531\u6b64\u53ef\u63a8\u7b97\u51fa \u570b\u8a9e\u97fb\u6bcd\u53ef\u80fd\u7d44\u5408\u51fa\u7684\u6587\u8108\u6578\u91cf\u662f(11+6+1) \u00d7 37 \u00d7 (9+6+1)=10,",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "\u7dd2\u8ad6",
                "sec_num": "1."
            }
        ],
        "back_matter": [],
        "bib_entries": {
            "BIBREF0": {
                "ref_id": "b0",
                "title": "Regularization techniques for discrete cepstrum estimation",
                "authors": [
                    {
                        "first": "O",
                        "middle": [],
                        "last": "Capp\u00e9",
                        "suffix": ""
                    },
                    {
                        "first": "&",
                        "middle": [
                            "E"
                        ],
                        "last": "Moulines",
                        "suffix": ""
                    }
                ],
                "year": 1996,
                "venue": "IEEE Signal Processing Letters",
                "volume": "3",
                "issue": "4",
                "pages": "100--102",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Capp\u00e9, O., & E. Moulines (1996). Regularization techniques for discrete cepstrum estimation. IEEE Signal Processing Letters, 3(4), 100-102.",
                "links": null
            },
            "BIBREF1": {
                "ref_id": "b1",
                "title": "A discrete-cepstrum based spectrum envelope estimation scheme and its example application of voice transformation",
                "authors": [
                    {
                        "first": "H",
                        "middle": [
                            "Y"
                        ],
                        "last": "Gu",
                        "suffix": ""
                    },
                    {
                        "first": "S",
                        "middle": [
                            "F"
                        ],
                        "last": "Tsai",
                        "suffix": ""
                    }
                ],
                "year": 2009,
                "venue": "International Journal of Computational Linguistics and Chinese Language Processing",
                "volume": "14",
                "issue": "4",
                "pages": "363--382",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Gu, H. Y., & S. F. Tsai (2009). A discrete-cepstrum based spectrum envelope estimation scheme and its example application of voice transformation. International Journal of Computational Linguistics and Chinese Language Processing, 14(4), 363-382.",
                "links": null
            },
            "BIBREF2": {
                "ref_id": "b2",
                "title": "Model spectrum-progression with DTW and ANN for speech synthesis",
                "authors": [
                    {
                        "first": "H",
                        "middle": [
                            "Y"
                        ],
                        "last": "Gu",
                        "suffix": ""
                    },
                    {
                        "first": "C",
                        "middle": [
                            "Y"
                        ],
                        "last": "Wu",
                        "suffix": ""
                    }
                ],
                "year": 2009,
                "venue": "Proc. ECTI-CON",
                "volume": "",
                "issue": "",
                "pages": "1010--1013",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Gu, H. Y., & C. Y. Wu (2009). Model spectrum-progression with DTW and ANN for speech synthesis. In Proc. ECTI-CON, Pattaya, Thailand, 1010-1013.",
                "links": null
            },
            "BIBREF3": {
                "ref_id": "b3",
                "title": "Combining HMM spectrum models and ANN prosody models for speech synthesis of syllable prominent languages",
                "authors": [
                    {
                        "first": "H",
                        "middle": [
                            "Y"
                        ],
                        "last": "Gu",
                        "suffix": ""
                    },
                    {
                        "first": "M",
                        "middle": [
                            "Y"
                        ],
                        "last": "Lai",
                        "suffix": ""
                    },
                    {
                        "first": "&",
                        "middle": [
                            "S F"
                        ],
                        "last": "Tsai",
                        "suffix": ""
                    }
                ],
                "year": 2010,
                "venue": "Proc. ISCSLP",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Gu, H. Y., M. Y. Lai, & S. F. Tsai (2010). Combining HMM spectrum models and ANN prosody models for speech synthesis of syllable prominent languages. In Proc. ISCSLP, Tainan, Taiwan, Special Session 1.",
                "links": null
            },
            "BIBREF4": {
                "ref_id": "b4",
                "title": "Integrating speaker-nonspecific timbre transformation to an HNM based speech synthesis scheme",
                "authors": [
                    {
                        "first": "H",
                        "middle": [
                            "Y"
                        ],
                        "last": "Gu",
                        "suffix": ""
                    },
                    {
                        "first": "C",
                        "middle": [
                            "L"
                        ],
                        "last": "Tsai",
                        "suffix": ""
                    }
                ],
                "year": 2013,
                "venue": "Journal of the Chinese Institute of Engineers",
                "volume": "36",
                "issue": "3",
                "pages": "371--381",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Gu, H. Y., & C. L. Tsai (2013). Integrating speaker-nonspecific timbre transformation to an HNM based speech synthesis scheme. Journal of the Chinese Institute of Engineers, 36(3), 371-381.",
                "links": null
            },
            "BIBREF5": {
                "ref_id": "b5",
                "title": "Exploiting prosody hierarchy and dynamic features for pitch modeling and generation in HMM-based speech synthesis",
                "authors": [
                    {
                        "first": "C",
                        "middle": [
                            "C"
                        ],
                        "last": "Hsia",
                        "suffix": ""
                    },
                    {
                        "first": "C",
                        "middle": [
                            "H"
                        ],
                        "last": "Wu",
                        "suffix": ""
                    },
                    {
                        "first": "&",
                        "middle": [
                            "J Y"
                        ],
                        "last": "Wu",
                        "suffix": ""
                    }
                ],
                "year": 2010,
                "venue": "IEEE trans. Audio, Speech, and Language Processing",
                "volume": "18",
                "issue": "8",
                "pages": "1994--2003",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Hsia, C. C., C. H. Wu, & J. Y. Wu (2010). Exploiting prosody hierarchy and dynamic features for pitch modeling and generation in HMM-based speech synthesis. IEEE trans. Audio, Speech, and Language Processing, 18(8), 1994-2003.",
                "links": null
            },
            "BIBREF6": {
                "ref_id": "b6",
                "title": "Fundamentals of Speech Recognition",
                "authors": [
                    {
                        "first": "L",
                        "middle": [
                            "B H"
                        ],
                        "last": "Rabiner",
                        "suffix": ""
                    },
                    {
                        "first": "",
                        "middle": [],
                        "last": "Juang",
                        "suffix": ""
                    }
                ],
                "year": 1993,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Rabiner, L. & B. H. Juang (1993). Fundamentals of Speech Recognition, Prentice Hall.",
                "links": null
            },
            "BIBREF7": {
                "ref_id": "b7",
                "title": "Speech parameter generation algorithm considering global variance for HMM-based speech synthesis",
                "authors": [
                    {
                        "first": "T",
                        "middle": [],
                        "last": "Toda",
                        "suffix": ""
                    },
                    {
                        "first": "&",
                        "middle": [
                            "K"
                        ],
                        "last": "Tokuda",
                        "suffix": ""
                    }
                ],
                "year": 2005,
                "venue": "Proc. Eurospeech",
                "volume": "",
                "issue": "",
                "pages": "2801--2804",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Toda, T., & K. Tokuda (2005). Speech parameter generation algorithm considering global variance for HMM-based speech synthesis. In Proc. Eurospeech, Lisbon, Portugal, 2801-2804.",
                "links": null
            },
            "BIBREF8": {
                "ref_id": "b8",
                "title": "An HMM-based approach to multilingual speech synthesis",
                "authors": [
                    {
                        "first": "K",
                        "middle": [],
                        "last": "Tokuda",
                        "suffix": ""
                    },
                    {
                        "first": "H",
                        "middle": [],
                        "last": "Zen",
                        "suffix": ""
                    },
                    {
                        "first": "&",
                        "middle": [
                            "A W"
                        ],
                        "last": "Black ; Narayanan",
                        "suffix": ""
                    },
                    {
                        "first": "A",
                        "middle": [],
                        "last": "Alwan",
                        "suffix": ""
                    }
                ],
                "year": 2004,
                "venue": "Text to Speech Synthesis: New Paradigms and Advances",
                "volume": "",
                "issue": "",
                "pages": "135--153",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Tokuda, K., H. Zen, & A. W. Black (2004). An HMM-based approach to multilingual speech synthesis. In Text to Speech Synthesis: New Paradigms and Advances, Editors: S. Narayanan and A. Alwan, Prentice Hall, 135-153.",
                "links": null
            },
            "BIBREF9": {
                "ref_id": "b9",
                "title": "Rich context modeling for high quality HMM-based TTS",
                "authors": [
                    {
                        "first": "Z",
                        "middle": [
                            "J"
                        ],
                        "last": "Yan",
                        "suffix": ""
                    },
                    {
                        "first": "Y",
                        "middle": [],
                        "last": "Qian",
                        "suffix": ""
                    },
                    {
                        "first": "&",
                        "middle": [
                            "F K"
                        ],
                        "last": "Soong",
                        "suffix": ""
                    }
                ],
                "year": 2009,
                "venue": "Proc. INTERSPEECH",
                "volume": "",
                "issue": "",
                "pages": "1755--1758",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Yan, Z. J., Y. Qian, & F. K. Soong (2009). Rich context modeling for high quality HMM-based TTS. In Proc. INTERSPEECH, Brighton, UK, 1755-1758.",
                "links": null
            },
            "BIBREF10": {
                "ref_id": "b10",
                "title": "Simultaneous modeling of spectrum, pitch and duration in HMM based speech synthesis",
                "authors": [
                    {
                        "first": "T",
                        "middle": [],
                        "last": "Yoshimura",
                        "suffix": ""
                    },
                    {
                        "first": "K",
                        "middle": [],
                        "last": "Tokuda",
                        "suffix": ""
                    },
                    {
                        "first": "T",
                        "middle": [],
                        "last": "Masuko",
                        "suffix": ""
                    },
                    {
                        "first": "T",
                        "middle": [],
                        "last": "Kobayashi",
                        "suffix": ""
                    },
                    {
                        "first": "&",
                        "middle": [
                            "T"
                        ],
                        "last": "Kitamura",
                        "suffix": ""
                    }
                ],
                "year": 1999,
                "venue": "Proc. Eurospeech",
                "volume": "",
                "issue": "",
                "pages": "2347--2350",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Yoshimura, T., K. Tokuda, T. Masuko, T. Kobayashi, & T. Kitamura (1999). Simultaneous modeling of spectrum, pitch and duration in HMM based speech synthesis. In Proc. Eurospeech, Budapest, Hungary, 2347-2350.",
                "links": null
            },
            "BIBREF11": {
                "ref_id": "b11",
                "title": "The HMM-based speech synthesis system (HTS) version 2.0",
                "authors": [
                    {
                        "first": "H",
                        "middle": [],
                        "last": "Zen",
                        "suffix": ""
                    },
                    {
                        "first": "T",
                        "middle": [],
                        "last": "Nose",
                        "suffix": ""
                    },
                    {
                        "first": "J",
                        "middle": [],
                        "last": "Yamagishi",
                        "suffix": ""
                    },
                    {
                        "first": "S",
                        "middle": [],
                        "last": "Sako",
                        "suffix": ""
                    },
                    {
                        "first": "T",
                        "middle": [],
                        "last": "Masuko",
                        "suffix": ""
                    },
                    {
                        "first": "A",
                        "middle": [
                            "W"
                        ],
                        "last": "Black",
                        "suffix": ""
                    },
                    {
                        "first": "&",
                        "middle": [
                            "K"
                        ],
                        "last": "Tokuda",
                        "suffix": ""
                    }
                ],
                "year": 2007,
                "venue": "Proc. 6th ISCA Workshop on Speech Synthesis",
                "volume": "",
                "issue": "",
                "pages": "294--299",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Zen, H., T. Nose, J. Yamagishi, S. Sako, T. Masuko, A. W. Black, & K. Tokuda (2007). The HMM-based speech synthesis system (HTS) version 2.0. In Proc. 6th ISCA Workshop on Speech Synthesis, Bonn, Germany, 294-299.",
                "links": null
            }
        },
        "ref_entries": {
            "TABREF1": {
                "type_str": "table",
                "content": "<table><tr><td>\u6211\u5011\u9080\u8acb\u4e86\u4e00\u4f4d\u6210\u5e74\u7537\u6027\u81f3\u9694\u97f3\u9304\u97f3\u5ba4\u9304\u88fd 1,208 \u500b\u8a9e\u53e5\u7684\u767c\u97f3\uff0c\u9019\u4e9b\u8a9e\u53e5\u7684\u8173\u672c \u6bcd)\u7684\u5de6\u53f3\u6587\u8108\u4f5c\u5206\u985e\uff0c\u4ee5\u964d\u4f4e\u6587\u8108\u7d44\u5408\u4e4b\u6578\u91cf\uff1b\u6b64\u5916\uff0c\u66f4\u9032\u4e00\u6b65\u7814\u7a76\u63d0\u51fa\u6587\u8108\u76f8\u4f9d\u4e4b\u534a</td></tr><tr><td>(script)\u662f\u96a8\u6a5f\u5730\u5f9e\u7121\u95dc\u7684\u6587\u7ae0\u4e2d\u6311\u9078\u51fa\uff0c\u7e3d\u8a08\u6709 10,173 \u500b\u97f3\u7bc0\uff0c\u800c\u9304\u97f3\u7684\u53d6\u6a23\u7387\u70ba 22,050 \u6bb5\u5f0f HMM \u7d50\u69cb\uff0c\u4ee5\u4fbf\u5728\u6709\u9650\u8a9e\u6599\u7684\u60c5\u6cc1\u4e0b\uff0c\u638c\u63e1\u4e00\u500b\u8a9e\u97f3\u55ae\u5143\u7684\u6587\u8108\u76f8\u4f9d\u983b\u8b5c\u7279\u6027\u3002</td></tr><tr><td>Hz\u3002\u70ba\u4e86\u6a19\u8a18\u9019\u4e9b\u8a9e\u53e5\u4e2d\u5404\u97f3\u7bc0\u7684\u62fc\u97f3\u7b26\u865f\uff0c\u6211\u5011\u9996\u5148\u4f7f\u7528 HTK \u5957\u4ef6\u4f86\u4f5c forced \u5982\u6b64\u7d50\u5408\u5169\u8005\uff0c\u7528\u4ee5\u6539\u9032\u5408\u6210\u8a9e\u97f3\u7684\u6d41\u66a2\u5ea6\u3002</td></tr><tr><td>alignment \u8655\u7406\uff0c\u800c\u5f97\u5230\u521d\u6b65\u7684\u97f3\u7bc0\u908a\u754c\u4e4b\u6a19\u8a18\uff0c\u7136\u5f8c\u4ee5\u4eba\u5de5\u64cd\u4f5c WaveSurfer \u8edf\u9ad4\u53bb\u66f4 \u70ba\u4e86\u8a55\u4f30\u672c\u8ad6\u6587\u6240\u63d0\u51fa\u7684\u534a\u6bb5\u5f0f HMM \u7d50\u69cb\uff0c\u6211\u5011\u9032\u884c\u4e86\u5169\u7a2e\u5be6\u9a57\uff0c\u5373\u983b\u8b5c\u8ddd\u96e2\u91cf</td></tr><tr><td>\u6b63\u932f\u8aa4\u7684\u97f3\u7bc0\u908a\u754c\u6a19\u8a18\u3002 \u6e2c\u548c\u6d41\u66a2\u5ea6\u807d\u6e2c\uff0c\u91cf\u6e2c\u51fa\u7684\u5e73\u5747\u983b\u8b5c\u8ddd\u96e2\u986f\u793a\uff0c\u4f7f\u7528\u534a\u6bb5\u5f0f HMM \u7d50\u69cb\u6240\u5408\u6210\u51fa\u7684\u8a9e\u53e5\uff0c</td></tr><tr><td>\u5927\u5e45\u964d\u4f4e\u6240\u9700\u7684\u8a13\u7df4\u8a9e\u6599\u6578\u91cf\uff0c\u5982 HTS \u8edf\u9ad4\u5c31\u662f\u63a1\u53d6\u6b64\u7a2e\u4f5c\u6cd5\u3002\u4e0d\u904e\uff0c\u672c\u8ad6\u6587\u63a1\u53d6\u53e6\u5916 \u4e00\u7a2e\u7814\u7a76\u65b9\u5411(approach)\uff0c\u5c31\u662f\u5148\u4f9d\u64da\u767c\u97f3\u77e5\u8b58\u4f86\u5c0d\u8072\u3001\u97fb\u6bcd\u4f5c\u5206\u985e(\u9019\u76f8\u7576\u65bc\u5c0d HMM \u4f5c\u5206\u7fa4)\uff0c\u518d\u7814\u7a76\u65b0\u7684 HMM \u7d50\u69cb\u4e4b\u8a2d\u8a08\uff0c\u4ee5\u4fbf\u89e3\u6c7a\u524d\u8ff0\u7684\u9700\u6c42\u5927\u91cf\u8a13\u7df4\u8a9e\u6599\u7684\u554f\u984c\u3002 3. \u5efa\u69cb\u65b0\u7684HMM\u7d50\u69cb \u5982\u5716 3 4. \u5be6\u9a57\u8a55\u4f30 \u5728\u672c\u8ad6\u6587\u4e2d\uff0c\u6211\u5011\u61c9\u7528\u97f3\u7d20\u7684\u767c\u97f3\u77e5\u8b58\u65bc\u53d6\u4ee3\u6c7a\u7b56\u6a39\uff0c\u4f86\u5c0d\u4e00\u500b\u8a9e\u97f3\u55ae\u5143(\u8072\u6bcd\u6216\u97fb 4.1 \u8a13\u7df4\u968e\u6bb5 \u6211\u5011\u5c07\u5404\u500b\u97f3\u7bc0\u7684\u97f3\u6a94\u5207\u5272\u6210\u4e00\u5e8f\u5217\u7684\u97f3\u6846\uff0c\u97f3\u6846\u9577\u5ea6\u8a2d\u70ba 512 \u500b\u6a23\u672c\u9ede\uff0c\u800c\u97f3\u6846 \u4f4d\u79fb\u8a2d\u70ba 128 \u500b\u6a23\u672c\u9ede\u3002\u6bcf\u500b\u97f3\u6846\u7d93\u5206\u6790\u8a08\u7b97\u5f8c\u64f7\u53d6\u51fa 39 \u500b\u983b\u8b5c\u53c3\u6578 c 0 , c 1 , \u2026, c 38 \uff0c\u5be6 \u969b\u4e0a\u662f\u96e2\u6563\u5012\u983b\u8b5c\u4fc2\u6578(discrete cepstral coefficients, DCC) (Capp\u00e9 &amp; Moulines, 1996)\uff0c DCC \u4fc2\u6578\u7684\u64f7\u53d6\u65b9\u6cd5\uff0c\u8acb\u53c3\u8003\u6211\u5011\u5148\u524d\u7684\u7814\u7a76\u8ad6\u6587(Gu &amp; Tsai, 2009)\u3002\u6b64\u5916\uff0c\u4e00\u500b\u97f3\u6846 \u7684\u9031\u671f\u6027\u8cc7\u8a0a\u4e5f\u6703\u88ab\u8a18\u9304\u5728\u53e6\u4e00\u500b\u7dad\u5ea6\u4e2d\uff0c\u5373\u5b58\u5165 c 39 \uff0c\u5982\u679c\u4e00\u500b\u97f3\u6846\u88ab\u5075\u6e2c\u51fa\u662f\u9031\u671f\u6027 \u7684\uff0c\u5247\u8a2d\u5b9a c 39 \u4e4b\u8a13\u7df4\uff0c\u6211\u5011\u4f9d\u64da\u5206\u6bb5\u5f0f K \u4e2d\u5fc3(segmental K-means)\u6f14\u7b97\u6cd5(Rabiner &amp; Juang, 1993)\uff0c\u53bb \u767c\u5c55\u4e86\u8a13\u7df4\u7a0b\u5f0f\u3002 4.2 \u8a9e\u97f3\u5408\u6210\u8655\u7406 \u672c\u8ad6\u6587\u88fd\u4f5c\u7684\u570b\u8a9e\u8a9e\u97f3\u5408\u6210\u7cfb\u7d71\uff0c\u5176\u5408\u6210\u968e\u6bb5\u4e4b\u8655\u7406\u6d41\u7a0b\u5982\u5716 1 \u6240\u793a\uff0c\u5340\u584a(a)\u4f5c \u6587 \u53e5\u5206\u6790(text analysis)\uff0c\u6bcf\u6b21\u6703\u5f9e\u8f38\u5165\u7684\u6a94\u6848\u8b80\u53d6\u4e00\u500b\u6587\u53e5\u9032\u4f86\uff0c\u7136\u5f8c\u4ee5\u67e5\u8a5e\u5178\u53ca\u6aa2\u67e5\u6578 \u500b\u69cb\u8a5e\u898f\u5247\u7684\u65b9\u5f0f\u53bb\u4f5c\u5256\u6790\uff0c\u628a\u8b80\u5165\u7684\u6587\u53e5\u5207\u5272\u6210\u4e00\u5e8f\u5217\u7684\u8a5e\u8a9e(words)\uff0c\u4e26\u4e14\u6bcf\u4e00\u500b\u8a5e \u8a9e\u7d93\u7531\u67e5\u8a5e\u5178\u4e5f\u53ef\u5f97\u77e5\u5b83\u7684\u62fc\u97f3\u7b26\u865f\u3002\u63a5\u8457\uff0c\u5728\u5340\u584a(b)\u7522\u751f\u5404\u97f3\u7bc0\u4e4b\u97f3\u9ad8\u8ecc\u8de1(pitch contour)\u53ca\u6642\u9577(duration)\u503c\uff0c\u5c0d\u65bc\u6bcf\u4e00\u500b\u97f3\u7bc0\uff0c\u5148\u6e96\u5099\u597d\u5b83\u7684\u6587\u8108\u8cc7\u6599\u9805\uff0c\u518d\u5c07\u6587\u8108\u8cc7\u6599 \u8f38\u5165\u81f3\u5169\u500b\u985e\u795e\u7d93\u7db2\u8def(ANN)\uff0c\u4ee5\u5206\u5225\u9810\u6e2c\u51fa\u97f3\u9ad8\u8ecc\u8de1\u53c3\u6578\u548c\u6642\u9577\u53c3\u6578\u7684\u503c\uff0c\u95dc\u65bc ANN \u7684\u8f38\u51fa/\u8f38\u5165\u8cc7\u6599\u9805\u3001\u53ca\u7d50\u69cb\u7684\u7d30\u7bc0\uff0c\u8acb\u53c3\u8003\u6211\u5011\u5148\u524d\u7684\u7814\u7a76\u8ad6\u6587(Gu &amp; Wu, 2009)\u3002 \u5728\u5340\u584a(c)\u6311\u9078\u8072\u6bcd\u3001\u97fb\u6bcd\u4e4b HMM \u6a21\u578b\uff0c\u9996\u5148\u4f9d\u64da\u5404\u500b\u97f3\u7bc0\u7684\u62fc\u97f3\u7b26\u865f\u53bb\u67e5\u8a62\u51fa\u5c0d \u61c9\u7684\u8072\u6bcd\u548c\u97fb\u6bcd\u4e4b\u62fc\u97f3\u7b26\u865f\u8207\u7de8\u865f\uff0c\u4e26\u4e14\u6c7a\u5b9a\u8072\u3001\u97fb\u6bcd\u5728\u8868 1\u3001\u8868 2\u3001\u8207\u8868 3 \u7684\u5206\u985e\u7de8\u865f\uff0c \u7136\u5f8c\u4f9d\u64da\u5404\u500b\u55ae\u5143(\u8072\u6bcd\u6216\u97fb\u6bcd)\u7684\u7de8\u865f\u548c\u5176\u524d\u5f8c\u6587\u8108\u7684\u5206\u985e\u7de8\u865f\uff0c\u5f9e\u8a13\u7df4\u51fa\u7684\u534a\u6bb5\u5f0f HMM \u6a21\u578b\u4e2d\uff0c\u627e\u51fa\u4e00\u500b\u55ae\u5143(\u8072\u6bcd\u6216\u97fb\u6bcd)\u5c0d\u61c9\u7684\u524d\u3001\u5f8c\u5169\u500b\u534a\u6bb5\u5f0f HMM\uff0c\u63a5\u8457\u628a\u8072\u3001 \u4f86\u8a08\u7b97\u5404\u500b\u97f3\u6846\u7684 DCC \u4fc2\u6578\u3002 \u5728\u5340\u584a(f)\u8a08\u7b97\u5404\u500b\u97f3\u6846\u7684\u97f3\u9ad8\u503c\uff0c\u4e00\u500b\u6709\u8072\u55ae\u5143(\u4f8b\u5982\u8072\u6bcd/m/\u548c\u97fb\u6bcd/a/)\u7684\u5404\u500b\u97f3\u6846 \u90fd\u5fc5\u9808\u88ab\u6307\u6d3e\u4e00\u500b\u97f3\u9ad8\u983b\u7387\u503c(\u55ae\u4f4d Hz)\uff0c\u5728\u6b64\u6211\u5011\u62ff ANN \u7522\u751f\u7684\u57fa\u9031\u8ecc\u8de1\u53c3\u6578\u53bb\u4f5c\u62c9 \u683c\u862d\u65e5\u5167\u63d2(Lagrange interpolation)\uff0c\u4ee5\u6c42\u5f97\u5404\u500b\u97f3\u6846\u7684\u97f3\u9ad8\u983b\u7387\u503c\u3002\u63a5\u8457\uff0c\u5728\u5340\u584a(g)\u63a1 \u7528 HNM \u4fe1\u865f\u6a21\u578b\u4f5c\u8a9e\u97f3\u4fe1\u865f\u5408\u6210\uff0c\u6211\u5011\u628a\u4e00\u500b\u55ae\u5143\u5404\u97f3\u6846\u7684 DCC \u4fc2\u6578\u8207\u97f3\u9ad8\u983b\u7387\u503c\uff0c \u6309\u7167\u97f3\u6846\u6b21\u5e8f\u9010\u500b\u97f3\u6846\u9001\u7d66 HNM \u8a9e\u97f3\u4fe1\u865f\u5408\u6210\u6a21\u7d44\uff0c\u53bb\u5408\u6210\u51fa\u8a9e\u97f3\u4fe1\u865f\uff0c\u95dc\u65bc HNM \u4fe1 \u865f\u5408\u6210\u4e4b\u7d30\u90e8\u8655\u7406\u65b9\u6cd5\uff0c\u53ef\u53c3\u8003\u6211\u5011\u5148\u524d\u7684\u7814\u7a76\u8ad6\u6587(Gu &amp; Tsai, 2013; Gu &amp; Tsai, 2009)\u3002 4.3 \u983b\u8b5c\u8ddd\u96e2\u91cf\u6e2c \u5728\u6b64\u4ee5\u4ee3\u865f SYC \u8207 SYD \u8868\u793a\u672c\u7814\u7a76\u6240\u5efa\u9020\u7684\u5169\u500b\u570b\u8a9e\u8a9e\u97f3\u5408\u6210\u7cfb\u7d71\uff0cSYC \u8868\u793a\u5b8c \u6574\u7684\u7cfb\u7d71\uff0c\u5c0d\u65bc\u4e00\u500b\u97f3\u7bc0\u7684\u524d\u63a5\u8207\u5f8c\u63a5\u6587\u8108\u90fd\u7d0d\u5165\u8003\u616e\uff0c\u9019\u610f\u5473\u8072\u6bcd\u7684\u524d\u534a\u6bb5 HMM \u7684 \u5de6\u6587\u8108\u6709\u4f5c\u5340\u5206(\u4f8b\u5982\u5716 4 \u4e4b GY_X)\uff0c\u4e26\u4e14\u97fb\u6bcd\u7684\u5f8c\u534a\u6bb5 HMM \u7684\u53f3\u6587\u8108\u4e5f\u6709\u4f5c\u5340\u5206(\u5982 \u5716 5 \u4e4b HY_Z)\uff0c\u6240\u4ee5\u5c0d\u65bc\u5404\u500b\u4e0d\u540c\u7684\u6587\u8108\u6a23\u5f0f\u5206\u985e(\u5982 CX)\uff0c\u5c31\u9700\u8981\u53bb\u5efa\u9020\u4e00\u500b\u5c0d\u61c9\u7684\u534a \u6bb5\u5f0f HMM (\u5982 GY_X)\u3002\u76f8\u53cd\u5730\uff0c\u5728\u7c21\u5316\u7684\u7cfb\u7d71 SYD \u88e1\uff0c\u5c0d\u65bc\u4e00\u500b\u97f3\u7bc0\u7684\u524d\u63a5\u8207\u5f8c\u63a5\u6587 \u8108\u5c31\u4e0d\u53bb\u4f5c\u5340\u5206\u4e86\uff0c\u4ea6\u5373\u8072\u6bcd\u7684\u524d\u534a\u6bb5 HMM \u4e0d\u53bb\u5340\u5206\u5b83\u7684\u5de6\u6587\u8108\uff0c\u5982\u6b64\u4e00\u500b\u8072\u6bcd\u5c31\u50c5 \u9700\u8a13\u7df4\u51fa\u4e00\u500b\u524d\u534a\u6bb5\u4e4b HMM\uff0c\u7136\u800c\u8072\u6bcd\u5f8c\u534a\u6bb5\u4e4b HMM\uff0c\u5247\u4ecd\u7136\u9700\u8a13\u7df4\u51fa\u6578\u500b\u534a\u6bb5\u5f0f HMM\uff0c\u4ee5\u5340\u5206\u53f3\u6587\u8108\uff1b\u985e\u4f3c\u9053\u7406\uff0c\u97fb\u6bcd\u7684\u5f8c\u534a\u6bb5 HMM \u5c31\u4e0d\u53bb\u5340\u5206\u5b83\u7684\u53f3\u6587\u8108\uff0c\u5982\u6b64\u4e00 \u500b\u97fb\u6bcd\u5c31\u53ea\u9700\u8a13\u7df4\u51fa\u4e00\u500b\u5f8c\u534a\u6bb5\u4e4b HMM\uff0c\u7136\u800c\u97fb\u6bcd\u524d\u534a\u6bb5\u4e4b HMM\uff0c\u5247\u4ecd\u7136\u9700\u8a13\u7df4\u51fa\u6578 \u500b\u534a\u6bb5\u5f0f HMM\uff0c\u4ee5\u5340\u5206\u5de6\u6587\u8108\u3002\u6b64\u5916\uff0c\u4ee5 SYP \u8868\u793a\u5148\u524d\u7814\u7a76\u88e1\u6240\u5efa\u9020\u7684\u570b\u8a9e\u8a9e\u97f3\u5408\u6210 \u7cfb\u7d71(Gu et al., 2010)\uff0c\u5728 SYP \u7cfb\u7d71\u88e1\uff0c\u6211\u5011\u5c0d\u65bc\u6bcf\u4e00\u7a2e\u570b\u8a9e\u97f3\u7bc0\u90fd\u5efa\u9020\u4e86\u4e00\u500b\u6216\u6578\u500b\u97f3 \u7bc0\u5bec\u5ea6\u7684 HMM\uff0c\u81f3\u65bc\u4e00\u7a2e\u97f3\u7bc0\u6240\u5efa\u9020\u7684 HMM \u500b\u6578\uff0c\u5247\u548c\u8a72\u97f3\u7bc0\u5728\u8a13\u7df4\u8a9e\u53e5\u4e2d\u7684\u767c\u97f3 \u6b21\u6578\u6709\u95dc\u3002\u5728\u672c\u7814\u88e1\uff0c\u6211\u5011\u4f7f\u7528\u76f8\u540c\u7684\u8a13\u7df4\u8a9e\u53e5\uff0c\u4f86\u8a13\u7df4\u9019\u4e09\u500b\u7cfb\u7d71: SYC\u3001SYD \u548c SYP\u3002 \u85c9\u7531\u9019\u4e09\u500b\u7cfb\u7d71\uff0c\u6211\u5011\u5c31\u53ef\u4ee5\u628a\u9304\u97f3\u8a9e\u53e5\u8207\u5408\u6210\u8a9e\u53e5\u76f8\u5c0d\u61c9\u7684\u97f3\u6846\u62ff\u53bb\u4f5c\u983b\u8b5c\u8ddd\u96e2 \u7684\u8a08\u7b97\u3002\u8a73\u7d30\u60c5\u5f62\u662f\uff0c\u628a 50 \u53e5\u6e2c\u8a66\u8a9e\u53e5\u7684\u6a19\u8a18\u6a94\u9010\u4e00\u8f38\u5165\u7d66 SYC\u3001SYD \u548c SYP \u7cfb\u7d71\u53bb \u8655\u7406\uff0c\u4ee5\u53d6\u5f97\u4e09\u500b\u7cfb\u7d71\u5c0d\u61c9\u65bc\u5404\u53e5\u6e2c\u8a66\u8a9e\u53e5\u7684\u7684\u5408\u6210\u97f3\u6a94\uff0c\u7136\u5f8c\u5206\u5225\u62ff\u5404\u53e5\u7684\u9304\u97f3\u8a9e\u53e5 \u53bb\u548c\u5408\u6210\u8a9e\u53e5\u53bb\u4f5c DCC \u5206\u6790\uff0c\u4ee5\u8a08\u7b97\u51fa\u5169\u500b DCC \u983b\u8b5c\u7279\u5fb5\u5411\u91cf\u7684\u5e8f\u5217\u3002\u63a5\u8457\uff0c\u5075\u6e2c\u5169 DCC \u5e8f\u5217\u4e2d\u5404\u7d44\u5c0d\u61c9\u97f3\u6846\u662f\u5426\u90fd\u70ba\u6709\u8072\uff0c\u82e5\u5c0d\u61c9\u7684\u97f3\u6846\u90fd\u5075\u6e2c\u70ba\u6709\u8072\uff0c\u5c31\u62ff\u8a72\u7d44\u97f3\u6846\u53bb \u8a08\u7b97\u97f3\u6846\u4e4b\u9593\u7684 DCC \u5411\u91cf\u5e7e\u4f55\u8ddd\u96e2\uff0c\u7136\u5f8c\u6211\u5011\u62ff 50 \u53e5\u6e2c\u8a66\u8a9e\u53e5\u7684\u6240\u6709\u6709\u8072\u97f3\u6846\u7b97\u51fa\u7684 \u5e7e\u4f55\u8ddd\u96e2\uff0c\u8a08\u7b97\u51fa\u4e00\u500b\u8de8\u8a9e\u53e5\u7684\u5e73\u5747\u8ddd\u96e2\u3002 SYC SYD SYP Avg. dist. 0.633 0.640 0.732 4.4 \u4e3b\u89c0\u807d\u6e2c\u5be6\u9a57 \u807d\u6e2c\u5be6\u9a57\u4f7f\u7528\u4e00\u7bc7\u8a13\u7df4\u8a9e\u53e5\u6c92\u6709\u7528\u5230\u7684\u77ed\u6587\uff0c\u8a72\u77ed\u6587\u5305\u62ec 70 \u500b\u97f3\u7bc0\uff0c\u6211\u5011\u5c07\u5b83\u5206 \u5225\u8f38\u5165\u5230\u4e09\u500b\u7cfb\u7d71\u53bb\u5408\u6210\u51fa\u8a9e\u97f3\u97f3\u6a94\uff0c\u5728\u6b64\u5206\u5225\u4ee5 WC\u3001WD \u548c WP \u4f86\u8868\u793a SYC\u3001SYD \u548c SYP \u9019 \u4e09 \u500b \u7cfb \u7d71 \u6240 \u5408 \u6210 \u7684 \u97f3 \u6a94 \uff0c \u9019 \u4e9b \u97f3 \u6a94 \u53ef \u5230 \u5982 \u4e0b \u7684 \u7db2 \u5740 \u53bb \u4e0b \u8f09 \u8207 \u8a66 \u807d : \u8005(\u524d\u8005)\u6bd4\u524d\u8005(\u5f8c\u8005)\u6d41\u66a2\u5f88\u591a\uff0c1(-1) \u5206\u8868\u793a\u5f8c\u8005(\u524d\u8005)\u6bd4\u524d\u8005(\u5f8c\u8005)\u7a0d\u5fae\u6d41\u66a2\uff0c0 \u5206\u8868 \u793a\u5206\u8fa8\u4e0d\u51fa\u4f86\u3002 \u4e09\u6b21\u807d\u6e2c\u5be6\u9a57\u4e4b\u5f8c\uff0c\u6211\u5011\u4f9d\u97f3\u6a94\u64ad\u653e\u6b21\u5e8f\u4f86\u8abf\u6574\u5206\u6578\u7684\u6b63\u8ca0\u865f\uff0c\u7136\u5f8c\u8a08\u7b97\u51fa\u5404\u6b21\u5be6 \u9a57\u7684\u5e73\u5747\u5206\u6578\u548c\u6a19\u6e96\u5dee\uff0c\u7d50\u679c\u5f97\u5230\u5982\u8868 5 \u6240\u793a\u7684\u6578\u503c\u3002\u5f9e\u8868 5 \u53ef\u77e5\u7b2c\u4e00\u6b21\u548c\u7b2c\u4e8c\u6b21\u5be6\u9a57 \u7684\u5e73\u5747\u5206\u6578\u70ba-0.833 \u548c-0.WC vs. WD WC vs. WP WD vs. WP AVG -0.833 -0.417 0.250 STD 0.718 0.900 0.866 5. \u7d50\u8a9e \u548c\u539f\u59cb\u9304\u97f3\u8a9e\u53e5\u4e4b\u9593\u7684\u983b\u8b5c\u8ddd\u96e2\uff0c\u53ef\u5f9e 0.732 \u6e1b\u5c11\u5230 0.633\uff1b\u6b64\u5916\uff0c\u807d\u6e2c\u5be6\u9a57\u7684\u7d50\u679c\u986f\u793a\uff0c \u4f7f\u7528\u534a\u6bb5\u5f0f HMM \u6240\u5408\u6210\u51fa\u7684\u8a9e\u97f3\uff0c\u6bd4\u4f7f\u7528\u53e6\u5916\u5169\u7a2e HMM \u7d50\u69cb\u7684\u8f03\u70ba\u6d41\u66a2\uff0c\u6240\u4ee5\u5728\u8a13 \u7df4\u8a9e\u53e5\u4e0d\u5145\u8db3\u7684\u60c5\u6cc1\u4e0b\uff0c\u534a\u6bb5\u5f0f HMM \u7d50\u69cb\u78ba\u5be6\u53ef\u6539\u9032\u5408\u6210\u8a9e\u97f3\u7684\u6d41\u66a2\u5ea6\u3002 \u672a\u4f86\u6211\u5011\u53ef\u5728\u76f8\u540c\u8a13\u7df4\u8a9e\u6599\u7684\u60c5\u6cc1\u4e0b\uff0c\u6bd4\u8f03\u6211\u5011\u7cfb\u7d71\u7684\u5408\u6210\u8a9e\u97f3\u8207 HTS \u8edf\u9ad4\u7684\u5408\u6210 \u8a9e\u97f3\uff0c\u89c0\u5bdf\u5169\u8005\u5728\u5ba2\u89c0\u983b\u8b5c\u8ddd\u96e2\u548c\u4e3b\u89c0\u807d\u6e2c\u4e0a\u7684\u5dee\u7570\u3002\u53e6\u5916\uff0c\u672c\u8ad6\u6587\u8457\u91cd\u65bc\u6539\u9032\u8a9e\u97f3\u55ae \u5143\u4e4b\u9593\u983b\u8b5c\u929c\u63a5\u4e0a\u7684\u6d41\u66a2\u5ea6\uff0c\u672a\u4f86\u53ef\u518d\u8003\u616e\u53bb\u6539\u9032\u97fb\u5f8b\u65b9\u9762\u7684\u6d41\u66a2\u5ea6\uff0c\u4ee5\u66f4\u70ba\u63d0\u5347\u7cfb\u7d71 \u6574\u9ad4\u7684\u6d41\u66a2\u5ea6\u3002 \u8868 4 System \u81f4\u8b1d \u611f\u8b1d\u570b\u79d1\u6703\u8a08\u756b\u4e4b\u7d93\u8cbb\u652f\u63f4\uff0c\u570b\u79d1\u6703\u8a08\u756b\u7de8\u865f: NSC 102-2221-E-011-129\u3002 http:/\u5011\u8981\u6c42\u4ed6\u7d66\u4e00\u500b\u5206\u6578\u4f86\u986f\u793a\u6d41\u66a2\u5ea6\u7684\u807d\u6e2c\u7d50\u679c\uff0c\u8a55\u5206\u7684\u7bc4\u570d\u70ba-2 \u5230 2 \u5206\uff0c2(-2)\u5206\u8868\u793a\u5f8c \u53c3\u8003\u6587\u737b</td></tr></table>",
                "text": "656 \u500b\u3002 \u5982\u679c\u6211\u5011\u4f9d\u64da\u524d\u8ff0\u7684\u6587\u8108\u7d44\u5408\u65b9\u5f0f\u53bb\u5efa\u9020 HMM \u6a21\u578b\uff0c\u5247\u9700\u8981\u8a13\u7df4\u7684 HMM \u6a21\u578b\u6703 \u6709 2,016 + 10,656 \u500b\uff0c\u9019\u610f\u5473\u8457\u5728\u6e96\u5099\u8a13\u7df4\u8a9e\u6599\u6642\uff0c\u5fc5\u9808\u9304\u88fd\u6578\u500d\u65bc 10,656 \u500b\u97f3\u7bc0\u6578\u91cf \u7684\u97f3\u7bc0\u767c\u97f3\u3002\u7136\u800c\u6e96\u5099\u5927\u91cf\u7684\u8a13\u7df4\u8a9e\u6599\u9700\u8981\u4ed8\u51fa\u6602\u8cb4\u7684\u8cbb\u7528\uff0c\u9019\u6697\u793a\u4f7f\u7528\u524d\u8ff0\u4e4b\u6587\u8108\u76f8 \u4f9d HMM \u6a21\u578b\u662f\u4e0d\u5207\u5be6\u969b\u7684\u3002\u5c0d\u65bc\u6b64\u554f\u984c\u7684\u89e3\u6c7a\u8fa6\u6cd5\uff0c\u5148\u524d\u7814\u7a76\u8005\u63d0\u51fa\u4f7f\u7528\u6c7a\u7b56\u6a39 (decision tree)\u4f86\u5c0d HMM \u6a21\u578b\u4f5c\u5206\u7fa4\uff0c\u518d\u5c0d\u5404\u7fa4\u5206\u5225\u53bb\u8a13\u7df4\u4e00\u500b\u5171\u7528\u7684 HMM\uff0c\u5982\u6b64\u5c31\u53ef \u6240\u756b\u7684\u662f\u64c1\u6709 6 \u500b\u72c0\u614b\u4e4b\u97fb\u6bcd HMM \u7684\u539f\u672c\u7d50\u69cb\uff0c\u6b64\u7d50\u69cb\u8868\u793a\u8a72\u97fb\u6bcd HMM \u662f\u5de6\u53f3\u6587\u8108\u76f8\u4f9d\u7684\uff0c\u7b26\u865f FY_XY \u8868\u793a\u6b64 HMM \u662f\u97fb\u6bcd Y \u7684\u6a21\u578b(F \u8868\u793a\u5de6\u53f3\u6587\u8108\u76f8\u4f9d)\uff0c \u4e26\u4e14 CX \u8868\u793a\u97fb\u6bcd Y \u524d\u63a5\u7684\u6587\u8108\u6a23\u5f0f(context type)\uff0cCZ \u5247\u8868\u793a Y \u5f8c\u63a5\u7684\u6587\u8108\u6a23\u5f0f\u3002\u6839\u64da \u8868 1 \u548c\u8868 3 \u5f97\u77e5\u6587\u8108\u6a23\u5f0f CX \u5171\u6709 12+6=18 \u7a2e\uff1b\u76f8\u5c0d\u5730\uff0c\u5f9e\u8868 1(\u9664\u4e86/n/\u8207/ng/\u4e4b\u5916)\u53ca\u8868 3 \u5f97\u77e5\u6587\u8108\u6a23\u5f0f CZ \u5171\u6709 10+6=16 \u7a2e\u3002\u6b64\u5916\uff0c\u570b\u8a9e\u6709 37 \u7a2e\u97fb\u6bcd Y\uff0c\u6240\u4ee5\u570b\u8a9e\u97fb\u6bcd\u7684\u6587\u8108 \u76f8\u4f9d\u4e4b HMM \u6a21\u578b\u7e3d\u8a08\u591a\u9054 10,656 \u500b\u3002 \u5716 3. \u5de6\u53f3\u6587\u8108\u76f8\u4f9d\u4e4b HMM \u6a21\u578b\u7d50\u69cb \u70ba\u4e86\u6e1b\u5c11\u5be6\u4f5c\u4e0a\u9700\u8981\u6295\u5165\u7684\u8cbb\u7528\u8207\u4eba\u529b(\u4f8b\u5982\u6e96\u5099\u5927\u91cf\u7684\u8a13\u7df4\u8a9e\u6599)\uff0c\u56e0\u6b64\u6211\u5011\u5617\u8a66 \u5c0d\u5716 3 \u7684 HMM \u6a21\u578b FY_XZ \u53bb\u91cd\u4f5c\u7d50\u69cb\u5b89\u6392\u3002\u6211\u5011\u7684\u89e3\u6c7a\u65b9\u6cd5\u662f\uff0c\u5047\u8a2d\u5716 3 \u4e2d\u524d\u9762\u534a \u6578\u7684\u72c0\u614b(\u5373\u72c0\u614b 1\u30012\u30013)\u6703\u76f8\u4f9d\u65bc\u524d\u63a5\u6587\u8108 CX\uff0c\u4f46\u662f\u548c\u5f8c\u63a5\u6587\u8108 CZ \u4e0d\u76f8\u95dc\uff1b\u985e\u4f3c\u5730\uff0c \u6211\u5011\u4e5f\u5047\u8a2d\u5716 3 \u4e2d\u5f8c\u9762\u534a\u6578\u4e4b\u72c0\u614b(\u5373\u72c0\u614b 4\u30015\u30016)\u53ea\u76f8\u4f9d\u65bc\u5f8c\u63a5\u6587\u8108 CZ\uff0c\u4f46\u662f\u548c\u524d\u63a5 \u6587\u8108 CX \u4e0d\u76f8\u95dc\u3002\u6839\u64da\u524d\u8ff0\u7684\u5169\u9805\u5047\u8a2d\uff0c\u5716 3 \u88e1\u5de6\u53f3\u6587\u8108\u4f9d\u4e4b HMM \u6a21\u578b FY_XZ\uff0c\u5c31\u53ef \u4ee5\u88ab\u5206\u89e3\u6210\u5716 4 \u8207\u5716 5 \u4e4b\u534a\u6bb5\u5f0f HMM \u6a21\u578b GY_X \u548c HY_Z\uff0c\u4ea6\u5373\u6211\u5011\u8981\u4ee5\u534a\u6bb5\u5f0f HMM \u6a21\u578b GY_X \u548c HY_Z \u4e4b\u4e32\u63a5\u4f86\u53d6\u4ee3 HMM \u6a21\u578b FY_XZ\u3002 \u5716 4. \u97fb\u6bcd Y \u524d\u534a\u6bb5\u4e4b\u5de6\u6587\u8108\u76f8\u4f9d HMM \u7d50\u69cb \u5716 5. \u97fb\u6bcd Y \u5f8c\u534a\u6bb5\u4e4b\u53f3\u6587\u8108\u76f8\u4f9d HMM \u7d50\u69cb \u7531\u65bc\u4e00\u500b\u97fb\u6bcd\u524d\u63a5\u7684\u6587\u8108\u88ab\u5206\u985e\u6210 18 \u7a2e\u6587\u8108\u6a23\u5f0f\uff0c\u6240\u4ee5\u6211\u5011\u9700\u8981\u5efa\u7acb 18 \u500b\u534a\u6bb5\u5f0f HMM \u6a21\u578b GY_X1, GY_X2, \u2026, GY_X18 (\u5982\u5716 5 \u88e1\u5217\u51fa)\uff0c\u4f86\u638c\u63e1\u97fb\u6bcd Y \u7684\u524d\u534a\u6bb5\u90e8\u5206\uff0c \u7b26\u865f GY \u4e4b G \u8868\u793a\u534a\u6bb5\u5f0f HMM \u4e4b\u524d\u534a\u6bb5\u3002\u985e\u4f3c\u5730\uff0c\u4e00\u500b\u97fb\u6bcd\u5f8c\u63a5\u7684\u6587\u8108\u88ab\u5206\u985e\u6210 16 \u7a2e\u6587\u8108\u6a23\u5f0f\uff0c\u6240\u4ee5\u6211\u5011\u9700\u8981\u5efa\u7acb 16 \u500b\u534a\u6bb5\u5f0f HMM \u6a21\u578b HY_Z1, HY_Z2, \u2026, HY_Z16 (\u5982 \u5716 4 \u88e1\u5217\u51fa)\uff0c\u4f86\u638c\u63e1\u97fb\u6bcd Y \u7684\u5f8c\u534a\u6bb5\u90e8\u5206\uff0c\u7b26\u865f HY \u4e4b H \u8868\u793a\u534a\u6bb5\u5f0f HMM \u4e4b\u5f8c\u534a\u6bb5\u3002 \u5982\u6b64\uff0c\u4e00\u500b\u97fb\u6bcd\u9700\u8981\u5efa\u7acb\u7684\u534a\u6bb5\u5f0f HMM \u6a21\u578b\u6578\u91cf\u662f 18+16=34 \u500b\uff0c\u800c\u570b\u8a9e\u6709 37 \u7a2e\u97fb\u6bcd\uff0c \u6240\u4ee5\u7e3d\u5171\u9700\u8981\u5efa\u7acb\u7684\u534a\u6bb5\u5f0f HMM \u7684\u6578\u91cf\u662f\uff0c34 \u00d7 37 = 1,258 \u500b\uff0c1,258 \u6bd4\u8d77 10,656 \u500b\u5de6 \u53f3\u6587\u8108\u76f8\u4f9d\u4e4b\u97fb\u6bcd HMM \u5c11\u4e86\u8a31\u591a\u3002 \u95dc\u65bc\u570b\u8a9e 21 \u500b\u8072\u6bcd\u7684 HMM \u6a21\u578b\u7684\u5efa\u7acb\uff0c\u6211\u5011\u628a\u524d\u8ff0\u4e4b\u97fb\u6bcd HMM \u6a21\u578b\u7684\u5047\u8a2d\u5957 \u7528\u9032\u4f86\u3002\u7531\u65bc\u4e00\u500b\u8072\u6bcd\u524d\u63a5\u7684\u6587\u8108\u88ab\u5206\u985e\u6210 12 \u7a2e\u6587\u8108\u6a23\u5f0f\uff0c\u6240\u4ee5\u6211\u5011\u9700\u8981\u70ba\u4e00\u500b\u8072\u6bcd\u5efa \u7acb 12 \u500b\u534a\u6bb5\u5f0f HMM \u6a21\u578b\uff0c\u4f86\u638c\u63e1\u8a72\u8072\u6bcd\u7684\u524d\u534a\u6bb5\u90e8\u5206\uff1b\u6b64\u5916\uff0c\u4e00\u500b\u8072\u6bcd\u5f8c\u63a5\u7684\u6587\u8108\u88ab \u5206\u985e\u6210 8 \u7a2e\u6587\u8108\u6a23\u5f0f\uff0c\u6240\u4ee5\u6211\u5011\u9700\u8981\u70ba\u4e00\u500b\u8072\u6bcd\u5efa\u7acb 8 \u500b\u534a\u6bb5\u5f0f HMM \u6a21\u578b\uff0c\u4f86\u638c\u63e1\u8a72 \u8072\u6bcd\u7684\u5f8c\u534a\u6bb5\u90e8\u5206\u3002\u5982\u6b64\uff0c\u4e00\u500b\u8072\u6bcd\u9700\u8981\u5efa\u7acb\u7684\u534a\u6bb5\u5f0f HMM \u6a21\u578b\u6578\u91cf\u662f 12+8=20 \u500b\uff0c \u800c\u570b\u8a9e\u6709 21 \u7a2e\u8072\u6bcd\uff0c\u6240\u4ee5\u7e3d\u5171\u9700\u8981\u5efa\u7acb\u7684\u534a\u6bb5\u5f0f HMM \u7684\u6578\u91cf\u662f\uff0c20 \u00d7 21 = 420 \u500b\uff0c420 \u6bd4\u8d77 2,016 \u500b\u5de6\u53f3\u6587\u8108\u76f8\u4f9d\u4e4b\u8072\u6bcd HMM \u5c11\u4e86\u8a31\u591a\u3002 \u7684\u503c\u70ba 1\uff0c\u53cd\u4e4b\u5247\u8a2d\u5b9a c 39 \u7684\u503c\u70ba 0\u3002\u4e00\u500b HMM \u7d93\u904e\u8a13\u7df4\u4e4b\u5f8c\uff0c\u6211\u5011\u5c31\u53ef \u4ee5\u4f9d\u64da\u5e73\u5747\u5411\u91cf\u7684 c 39 \u7684\u503c\uff0c\u4f86\u5224\u65b7\u5404\u500b HMM \u72c0\u614b\u662f\u5426\u70ba\u6709\u8072(voiced)\u6216\u7121\u8072(unvoiced) \u4e4b\u72c0\u614b\uff0c\u7136\u5f8c\u5c31\u53ef\u70ba\u6709\u8072\u7684 HMM \u72c0\u614b\u53bb\u7522\u751f\u57fa\u9031\u8ecc\u8de1\u3002\u53e6\u5916\uff0c\u983b\u8b5c\u53c3\u6578\u7684\u5dee\u5206\u503c\u4e5f\u662f \u6709\u7528\u7684\uff0c\u6240\u4ee5\u6211\u5011\u628a\u983b\u8b5c\u7279\u5fb5\u5411\u91cf\u589e\u52a0 40 \u7dad\uff0c\u4ee5\u5132\u5b58 DCC \u4fc2\u6578\u7684\u4e00\u968e\u5dee\u5206\u503c\u3002\u5728\u8a13\u7df4 \u5b8c\u4e00\u500b HMM \u4e4b\u5f8c\uff0c\u9664\u4e86\u8a18\u9304 HMM \u7684\u6a21\u578b\u53c3\u6578\u4e4b\u5916\uff0c\u4e5f\u8981\u8a18\u9304\u5404\u500b HMM \u72c0\u614b\u88ab\u99d0\u7559 \u7684\u97f3\u6846\u500b\u6578\u4e4b\u5e73\u5747\u503c\u8207\u8b8a\u7570\u6578\u3002 \u5728\u672c\u8ad6\u6587\u4e2d\uff0c\u4e00\u500b\u97fb\u6bcd\u7684\u524d\u534a\u8207\u5f8c\u534a\u90e8\u5206\u4e4b\u534a\u6bb5\u5f0f HMM \u6a21\u578b\u7686\u662f\u7531 3 \u500b\u72c0\u614b\u5efa\u9020 \u800c\u6210\uff0c\u4e26\u4e14\u72c0\u614b\u79fb\u8f49\u65b9\u5f0f\u90fd\u662f\u7531\u5de6\u81f3\u53f3\uff0c\u5c31\u5982\u5716 4 \u548c\u5716 5 \u6240\u793a\u3002\u4e0d\u904e\uff0c\u5c0d\u65bc\u4e00\u500b\u8072\u6bcd\u7684 \u534a\u6bb5\u5f0f HMM \u6a21\u578b\uff0c\u6211\u5011\u50c5\u4f7f\u7528 2 \u500b\u72c0\u614b\u53bb\u5efa\u9020\u3002\u95dc\u65bc\u9ad8\u65af\u6df7\u5408\u7d44\u4ef6(Gaussian mixture component)\u7684\u6578\u91cf\uff0c\u5728\u6bcf\u4e00\u500b HMM \u72c0\u614b\u4e0a\uff0c\u6211\u5011\u53ea\u8a2d\u7f6e\u4e00\u500b\u9ad8\u65af\u6df7\u5408\u3002\u5c0d\u65bc\u534a\u6bb5\u5f0f HMM \u97fb\u6bcd\u7684\u56db\u500b\u534a\u6bb5\u5f0f HMM \u4f9d\u5e8f\u4e32\u63a5\u6210\u4e00\u500b\u97f3\u7bc0\u7684\u5b8c\u6574 HMM \u6a21\u578b\u3002\u5728\u5340\u584a(d)\u6c7a\u5b9a\u5404\u500b HMM \u72c0\u614b\u7684\u99d0\u7559\u97f3\u6846\u6578\uff0c\u6211\u5011\u63a1\u7528 Tokuda \u7b49\u4eba\u63d0\u51fa\u7684\u65b9\u6cd5(Tokuda et al., 2004)\uff0c\u4f9d\u64da ANN \u7522\u751f\u7684\u97f3\u7bc0\u6642\u9577\u503c\uff0c\u53bb\u8a08\u7b97\u4e00\u500b\u97f3\u7bc0 HMM \u4e4b\u5404\u500b\u72c0\u614b\u6240\u61c9\u5206\u914d\u5230\u7684\u6642\u9577\u97f3\u6846\u6578\u3002 \u63a5\u8457\uff0c\u5728\u5340\u584a(e)\u7522\u751f\u5404\u97f3\u6846\u4e4b\u983b\u8b5c\u4fc2\u6578(\u5373 DCC \u4fc2\u6578)\uff0c\u4e00\u500b\u5e38\u88ab\u4f7f\u7528\u7684\u65b9\u6cd5\u662f\u6700\u5927\u4f3c\u7136 \u4f30\u8a08\u6cd5(Maximum likelihood Estimate, MLE)(Tokuda et al., 2004)\uff0c\u4e0d\u904e\uff0c\u672c\u8ad6\u6587\u4f7f\u7528\u7684\u662f \u5148\u524d\u6211\u5011\u63d0\u51fa\u7684\u52a0\u6b0a\u5f0f\u7dda\u6027\u5167\u63d2\u6cd5(weighted linear interpolation, WLI)(Gu et al., 2010)\uff0c \u6240\u5217\u51fa\u7684\u6578\u503c\uff0c\u5c31\u662f\u9019\u4e09\u500b\u7cfb\u7d71\u7684\u5e73\u5747\u983b\u8b5c\u8ddd\u96e2\uff0c\u5f9e\u8868 4 \u53ef\u77e5 SYC \u7cfb\u7d71\u6240\u7522 \u751f\u7684\u97f3\u6846 DCC \u5411\u91cf\uff0c\u6700\u9760\u8fd1\u65bc\u9304\u97f3\u8a9e\u53e5\u5206\u6790\u51fa\u4f86\u7684 DCC \u5411\u91cf\uff0c\u800c SYP \u7cfb\u7d71\u6240\u7522\u751f\u7684\u97f3 \u6846 DCC \u5411\u91cf\uff0c\u6700\u9060\u96e2\u9304\u97f3\u8a9e\u53e5\u5206\u6790\u51fa\u4f86\u7684 DCC \u5411\u91cf\u3002\u6b64\u5916\uff0c\u53ea\u8981\u4e00\u500b\u97f3\u7bc0\u4e2d\u7684\u8072\u6bcd\u548c \u97fb\u6bcd\u4e4b\u9593\u7684\u6587\u8108\u76f8\u4f9d\u6027\u6709\u88ab\u638c\u63e1(modeled)\uff0c\u5373 SYD \u7cfb\u7d71\u7684\u60c5\u6cc1\uff0c\u5247\u91cf\u6e2c\u51fa\u7684 DCC \u983b\u8b5c \u8ddd\u96e2\uff0c\u5c31\u6703\u6bd4 SYP \u7cfb\u7d71\u7684\u597d\u5f88\u591a\uff0c\u9019\u8868\u793a\u5716 4 \u548c\u5716 5 \u6240\u5217\u51fa\u7684\u534a\u6bb5\u5f0f HMM \u7d50\u69cb\uff0c\u7684\u78ba \u53ef\u5e6b\u5fd9\u638c\u63e1\u5169\u500b\u76f8\u9130\u8a9e\u97f3\u55ae\u5143\u4e4b\u9593\u7684\u6587\u8108\u76f8\u4f9d\u4e4b\u983b\u8b5c\u7279\u6027\u3002 \u8868 4. \u5e73\u5747 DCC \u983b\u8b5c\u8ddd\u96e2 /guhy.csie.ntust.edu.tw/hmmhalf/\u3002 \u900f\u904e WC\u3001WD \u548c WP \u4e09\u500b\u97f3\u6a94\uff0c\u6211\u5011\u9032\u884c\u6d41\u66a2\u5ea6\u6bd4\u8f03\u7684\u807d\u6e2c\u5be6\u9a57\uff0c\u4e00\u5171\u9080\u8acb\u4e86 12 \u4f4d\u53d7\u6e2c\u8005\uff0c\u5728\u7b2c\u4e00\u6b21\u807d\u6e2c\u5be6\u9a57\u88e1\uff0c\u53d7\u6e2c\u8005\u4ee5\u96a8\u6a5f\u6b21\u5e8f\u8046\u807d WC \u548c WD \u97f3\u6a94\uff1b\u5728\u7b2c\u4e8c\u6b21\u807d \u6e2c\u5be6\u9a57\u88e1\uff0c\u53d7\u6e2c\u8005\u4ee5\u96a8\u6a5f\u6b21\u5e8f\u8046\u807d WC \u548c WP \u97f3\u6a94\uff1b\u5728\u7b2c\u4e09\u6b21\u807d\u6e2c\u5be6\u9a57\u88e1\uff0c\u53d7\u6e2c\u8005\u5247\u4ee5 \u96a8\u6a5f\u6b21\u5e8f\u8046\u807d WD \u548c WP \u97f3\u6a94\u3002\u5728\u5404\u6b21\u807d\u6e2c\u5be6\u9a57\u4e2d\uff0c\u7576\u4e00\u4f4d\u53d7\u6e2c\u8005\u807d\u5b8c\u5169\u500b\u97f3\u6a94\u5f8c\uff0c\u6211 417\uff0c\u8ca0\u7684\u5206\u6578\u8868\u793a\u97f3\u6a94 WC \u6bd4 WD \u548c WP \u8f03\u70ba\u6d41\u66a2\uff0c\u5982\u679c\u5168 \u90e8\u53d7\u6e2c\u8005\u90fd\u5177\u6709\u8a9e\u97f3\u5408\u6210\u7814\u7a76\u7684\u80cc\u666f\uff0c\u6211\u5011\u8a8d\u70ba\u5169\u8ca0\u5206\u7684\u7d55\u5c0d\u503c\u61c9\u6703\u66f4\u5927\uff0c\u6240\u4ee5\u534a\u6bb5\u5f0f HMM \u7d50\u69cb\u7684\u78ba\u53ef\u4ee5\u6709\u6548\u7684\u63d0\u5347\u5408\u6210\u8a9e\u97f3\u7684\u6d41\u66a2\u5ea6\u3002\u81f3\u65bc\u7b2c\u4e09\u6b21\u5be6\u9a57\u7684\u5e73\u5747\u5206\u6578 0.250\uff0c \u8a72\u5206\u6578\u7684\u7d55\u5c0d\u503c\u6700\u5c0f\uff0c\u6211\u5011\u89ba\u5f97\u9019\u8868\u793a WD \u548c WP \u97f3\u6a94\u7684\u6d41\u66a2\u5ea6\u61c9\u7121\u660e\u986f\u7684\u5dee\u7570\u3002 \u8868 5. \u807d\u6e2c\u5be6\u9a57\u4e4b\u5e73\u5747\u8a55\u5206",
                "num": null,
                "html": null
            }
        }
    }
}