| { | |
| "paper_id": "O07-1005", | |
| "header": { | |
| "generated_with": "S2ORC 1.0.0", | |
| "date_generated": "2023-01-19T08:07:56.106262Z" | |
| }, | |
| "title": "Cyberon Voice Commander \u591a\u570b\u8a9e\u8a00\u8a9e\u97f3\u547d\u4ee4\u7cfb\u7d71 \u591a\u570b\u8a9e\u8a00\u8a9e\u97f3\u547d\u4ee4\u7cfb\u7d71 \u591a\u570b\u8a9e\u8a00\u8a9e\u97f3\u547d\u4ee4\u7cfb\u7d71 \u591a\u570b\u8a9e\u8a00\u8a9e\u97f3\u547d\u4ee4\u7cfb\u7d71", | |
| "authors": [], | |
| "year": "", | |
| "venue": null, | |
| "identifiers": {}, | |
| "abstract": "", | |
| "pdf_parse": { | |
| "paper_id": "O07-1005", | |
| "_pdf_hash": "", | |
| "abstract": [], | |
| "body_text": [], | |
| "back_matter": [], | |
| "bib_entries": { | |
| "BIBREF0": { | |
| "ref_id": "b0", | |
| "title": "HP iPAQ 510 Voice Messenger series -overview and features", | |
| "authors": [], | |
| "year": null, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "HP iPAQ 510 Voice Messenger series -overview and features. Available: http://h10010.www1.hp.com/wwpc/us/en/sm/WF05a/215348-215348-64929-314903-33 52590-3360087.html", | |
| "links": null | |
| }, | |
| "BIBREF1": { | |
| "ref_id": "b1", | |
| "title": "HP iPaq 510 Voice Messenger", | |
| "authors": [ | |
| { | |
| "first": "Bonnie", | |
| "middle": [], | |
| "last": "Cha", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2007, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Bonnie Cha, HP iPaq 510 Voice Messenger, CNET Asia Review, February 2007. Available: http://asia.cnet.com/reviews/mobilephones/0,39051199,40151061p,00.htm", | |
| "links": null | |
| }, | |
| "BIBREF2": { | |
| "ref_id": "b2", | |
| "title": "Subphonetic Acoustic Modeling for Speaker-Independent Continuous Speech Recognition", | |
| "authors": [ | |
| { | |
| "first": "M", | |
| "middle": [ | |
| "Y" | |
| ], | |
| "last": "Hwang", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1993, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Hwang, M. Y., Subphonetic Acoustic Modeling for Speaker-Independent Continuous Speech Recognition, PhD thesis, CMU-CS-92-230, Carnegie Mellon University, 1993", | |
| "links": null | |
| }, | |
| "BIBREF3": { | |
| "ref_id": "b3", | |
| "title": "The Use of Context in Large Vocabulary Speech Recognition", | |
| "authors": [ | |
| { | |
| "first": "J", | |
| "middle": [ | |
| "J" | |
| ], | |
| "last": "Odell", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1995, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Odell, J.J., The Use of Context in Large Vocabulary Speech Recognition, PhD thesis, Cambridge University, 1995", | |
| "links": null | |
| }, | |
| "BIBREF4": { | |
| "ref_id": "b4", | |
| "title": "Fundamentals of Speech Recognition", | |
| "authors": [ | |
| { | |
| "first": "L", | |
| "middle": [], | |
| "last": "Rabiner", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "B", | |
| "middle": [], | |
| "last": "Juang", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1993, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Rabiner, L. and Juang, B., Fundamentals of Speech Recognition. Prentice-Hall, 1993.", | |
| "links": null | |
| }, | |
| "BIBREF5": { | |
| "ref_id": "b5", | |
| "title": "Issues in building general letter to sound rules", | |
| "authors": [ | |
| { | |
| "first": "A", | |
| "middle": [], | |
| "last": "Black", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "K", | |
| "middle": [], | |
| "last": "Lenzo", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "V", | |
| "middle": [], | |
| "last": "Pagel", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1998, | |
| "venue": "ESCA Workshop on Speech Synthesis", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Black, A., Lenzo, K., and Pagel, V., \"Issues in building general letter to sound rules\", ESCA Workshop on Speech Synthesis, Jenolan Caves, Australia. 1998.", | |
| "links": null | |
| }, | |
| "BIBREF6": { | |
| "ref_id": "b6", | |
| "title": "Assigning Phrase Breaks from Part of Speech Sequences", | |
| "authors": [ | |
| { | |
| "first": "P", | |
| "middle": [], | |
| "last": "Taylor", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "A", | |
| "middle": [], | |
| "last": "Black", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1998, | |
| "venue": "Computer Speech and Language", | |
| "volume": "12", | |
| "issue": "", | |
| "pages": "99--117", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Taylor, P., and Black, A. \"Assigning Phrase Breaks from Part of Speech Sequences\", Computer Speech and Language. Vol. 12, pp. 99-117, 1998.", | |
| "links": null | |
| }, | |
| "BIBREF7": { | |
| "ref_id": "b7", | |
| "title": "Classification and Regression Trees", | |
| "authors": [ | |
| { | |
| "first": "L", | |
| "middle": [], | |
| "last": "Breiman", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "J", | |
| "middle": [ | |
| "H" | |
| ], | |
| "last": "Friedman", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "R", | |
| "middle": [ | |
| "A" | |
| ], | |
| "last": "Olshen", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "P", | |
| "middle": [ | |
| "J" | |
| ], | |
| "last": "Stone", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1984, | |
| "venue": "The Wadsworth Statistics/Probability Series", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Breiman, L., Friedman, J. H., Olshen, R. A., and Stone, P. J., Classification and Regression Trees. The Wadsworth Statistics/Probability Series, Wadsworth and Brooks, 1984.", | |
| "links": null | |
| } | |
| }, | |
| "ref_entries": { | |
| "TABREF0": { | |
| "num": null, | |
| "html": null, | |
| "type_str": "table", | |
| "text": "Fujitsu-Siemens Pocket Loox T830\u3001HTC Touch \u7b49\u591a\u6b3e\u66a2\u92b7\u624b\u6a5f\u90fd\u6709\u5167\u5efa\u591a\u570b\u8a9e\u8a00\u7248 CVC\u3002 \u64a5\u865f\u4e2d\u8acb\u7a0d\u5019 \u672c\u8ad6\u6587\u5c07\u63cf\u8ff0 CVC \u7684\u7cfb\u7d71\u67b6\u69cb\u4ee5\u53ca\u652f\u63f4\u591a\u570b\u8a9e\u8a00\u7684\u4f5c\u6cd5\uff0c\u7b2c\u4e8c\u7bc0\u5c07\u63cf\u8ff0 CVC \u7cfb \u7d71\u67b6\u69cb\uff0c\u5305\u62ec\u591a\u570b\u8a9e\u8a00\u7684\u8a9e\u97f3\u8fa8\u8b58\u548c TTS \u6240\u4f7f\u7528\u7684\u6280\u8853\uff0c\u7b2c\u4e09\u7bc0\u63cf\u8ff0\u8a9e\u97f3\u8fa8\u8b58\u5be6\u9a57\u74b0 \u5883\u548c\u5e7e\u500b\u8a9e\u8a00\u7248\u672c\u7684\u5be6\u9a57\u7d50\u679c\uff0c\u7b2c\u56db\u7bc0\u63d0\u51fa\u6211\u5011\u672a\u4f86\u5c0d CVC \u6539\u826f\u7684\u5e7e\u500b\u60f3\u6cd5\u3002 \u4e8c\u3001CVC \u7cfb\u7d71\u67b6\u69cb CVC \u5305\u62ec\u4e00\u7d44\u61c9\u7528\u7a0b\u5f0f\uff0c\u63d0\u4f9b\u4f7f\u7528\u8005\u4ecb\u9762\u3001\u9032\u884c\u9304\u653e\u97f3\u3001\u63a7\u5236\u6574\u500b\u548c\u4f7f\u7528\u8005\u5c0d\u8a71 \u7684\u6d41\u7a0b\uff0c\u5e95\u5c64\u70ba SIVR (speaker-independent voice recognition) \u6a21\u7d44\u548c TTS \u6a21\u7d44\uff0c\u5982\u5716\u4e8c \u6240\u793a\u3002SIVR \u6a21\u7d44\u53ef\u9032\u884c\u7368\u7acb\u8a5e\u5f59\u8fa8\u8b58\uff0c\u6216\u6709\u6587\u6cd5\u9650\u5236\u7684\u9023\u7e8c\u8a9e\u97f3\u547d\u4ee4\u8fa8\u8b58\uff0cTTS \u80fd\u5c07 \u8fa8\u8b58\u7684\u7d50\u679c\u6216\u751a\u81f3\u4efb\u610f\u6587\u5b57\u8f38\u5165\u8f49\u63db\u6210 PCM \u8072\u97f3\uff0c\u900f\u904e\u624b\u6a5f\u7684\u5587\u53ed\u64ad\u653e\u653e\u51fa\u4f86\u3002\u7531 \u65bc\u624b\u6a5f\u8cc7\u6e90\u8f03\u5c0f\uff0c\u8981\u80fd\u5728\u4e0a\u9762\u904b\u884c\uff0c\u6240\u6709\u7684\u904b\u7b97\u5fc5\u9808\u6539\u6210\u6574\u6578\u904b\u7b97\uff0c\u4e14\u4f7f\u7528\u7684\u7a7a\u9593\u4e5f\u4e0d \u5716\u4e00\u3001\u82f1\u6587\u7248 CVC \u756b\u9762 \u80fd\u592a\u5927\uff0c\u6211\u5011\u7684 SIVR \u548c TTS engine \u4e00\u500b\u8a9e\u8a00\u52a0\u8d77\u4f86\u5927\u7d04\u4f54\u7528 500KB \u5230 800KB\uff0c\u4f9d\u8a9e Windows Messenger \u300d \u662f \u7531 \u958b \u555f ( \u52d5 \u4f5c ) \u3001 Windows Messenger(\u61c9\u7528\u7a0b\u5f0f)\u6240\u7d44\u6210\u3002\u4ee5 CVC \u800c\u8a00\uff0c\u5176\u8a9e\u97f3\u547d\u4ee4\u7684\u6587\u6cd5\u9650\u5236\u5982\u5716\u56db\u6240\u793a\uff0c \u5176\u4e2d\u865b\u7dda\u4e0a\u7684\u8a5e\u5f59\u70ba\u975e\u5fc5\u8981\u7684\uff0c\u4f7f\u7528\u8005\u8b1b\u300c\u6253\u96fb\u8a71\u7d66\u4f55\u6cf0\u8ed2\u300d\u4e5f\u662f\u5408\u6cd5\u7684\u8a9e\u53e5\u3002\u5728 \u8fa8\u8b58\u7684\u904e\u7a0b\u4e2d\uff0cViterbi search \u9032\u884c word model \u4e4b\u9593\u7684\u72c0\u614b\u8f49\u79fb(state transition)\u6703\u53c3 \uff0c\u540c\u6642\u627e\u51fa\u5c0d\u61c9\u7684\u97f3\u6a19\u548c\u6bcf\u500b\u8a5e\u7684 POS (part-of-speech) tag\uff0c\u6709\u4e9b\u8a9e\u8a00\u9084\u9700 \u8981\u5148\u9032\u884c\u65b7\u8a5e\u8655\u7406\u3002\u6839\u64da\u6587\u53e5\u5206\u6790\u7d50\u679c\uff0c\u97fb\u5f8b\u6a21\u578b(prosodic model)\u5efa\u7acb\u6bcf\u500b\u7247\u8a9e\u7684\u97fb\u5f8b \u53c3\u6578\uff0c\u8a9e\u97f3\u5408\u6210\u5668(speech synthesizer)\u5247\u4e00\u6b21\u5408\u6210\u4e00\u500b\u7247\u8a9e\u7684\u8a9e\u97f3\uff1a\u7531\u8a9e\u97f3\u5408\u6210\u55ae\u4f4d\u8cc7 \u6599\u5eab(speech unit database)\u53d6\u5f97\u7247\u8a9e\u4e2d\u6240\u9700\u7684\u5408\u6210\u55ae\u5143\u8a9e\u97f3\u8cc7\u6599\uff0c\u4e26\u53c3\u8003\u97fb\u5f8b\u6a21\u578b\u63d0\u4f9b\u7684 \u8cc7\u8a0a\uff0c\u8abf\u6574\u7247\u8a9e\u7684\u97fb\u5f8b(prosody)\u8f38\u51fa\u5408\u6210\u8a9e\u97f3\u3002\u4ee5\u4e0b\u63cf\u8ff0 TTS \u7684\u5404\u500b\u7d44\u6210\u90e8\u5206\u904b\u4f5c\u904e \u7a0b\u3002 1. \u6587\u53e5\u5206\u6790 \u6574\u53e5\u6587\u5b57\u8f38\u5165\u5f8c\uff0c\u6211\u5011\u9808\u5148\u5c07\u6bcf\u500b\u8a5e\u6a19\u8a3b POS\uff0c\u7531\u65bc\u5728\u8a5e\u5178\u4e2d\u6bcf\u500b\u8a5e\u53ef\u80fd\u6709\u591a \u7a2e POS\uff0c\u6211\u5011\u4f7f\u7528 Viterbi search \u548c POS n-gram \u627e\u51fa\u6a5f\u7387\u6700\u9ad8\u7684 POS tag \u7d44\u5408\uff0cPOS n-gram \u662f\u4e8b\u5148\u7531\u4eba\u5de5\u6a19\u793a\u904e POS tag \u7684\u5927\u91cf\u6587\u5b57\u8cc7\u6599\u4e2d\u8a13\u7df4\u800c\u5f97\uff0c\u518d\u7531\u6b64\u6a19\u8a3b\u51fa\u7684 POS tag \u627e\u51fa\u6700\u53ef\u80fd\u7684\u7247\u8a9e\u908a\u754c(phrase boundary)\uff0c\u627e\u5c0b\u7684\u65b9\u6cd5\uff0c\u540c\u6a23\u5730\u4e5f\u662f\u4f7f\u7528 Viterbi search \u548c boundary n-gram \u627e\u51fa\u6a5f\u7387\u6700\u9ad8\u7684 boundary \u4f4d\u7f6e [7]\uff0cboundary n-gram \u4e5f\u662f\u4e8b\u5148\u7531\u4eba\u5de5\u6a19\u793a\u904e POS \u548c boundary tag \u7684\u5927\u91cf\u6587\u5b57\u8cc7\u6599\u4e2d\u8a13\u7df4\u800c\u5f97\u3002\u4f46 \u6709\u4e9b\u8a9e\u8a00\u6211\u5011\u7684\u5b57\u5178\u7f3a\u4e4f Linear Predictive Coding)\u3002\u6bcf\u500b\u8a9e\u8a00\u6211 \u5011\u627e\u4e00\u4f4d\u4ee5\u8a72\u8a9e\u8a00\u70ba\u6bcd\u8a9e\u7684\u5973\u6027\u8a9e\u8005\u9032\u884c\u9304\u97f3\uff0c\u5f9e\u6240\u9304\u88fd\u7684\u8a9e\u97f3\u4e2d\u5207\u5272\u51fa\u6240\u9700\u8981\u7684 \u8a9e\u97f3\u5408\u6210\u55ae\u5143(speech unit)\uff0c\u540c\u6642\u6c7a\u5b9a\u51fa\u6bcf\u500b\u8a9e\u97f3\u55ae\u4f4d\u7684 pitch \u4f4d\u7f6e\u3002\u6211\u5011\u5c0d\u6bcf\u500b pitch \u505a LPC \u5206\u6790\uff0c\u518d\u5c0d LPC \u4fc2\u6578\u548c residual \u58d3\u7e2e\u5132\u5b58\u65bc\u8cc7\u6599\u5eab\u4e2d\u3002\u5408\u6210\u6642\u97fb\u5f8b\u6a21\u578b (prosody model)\u6703\u9810\u6e2c\u51fa\u6574\u500b\u7247\u8a9e\u7684 F0 \u5f62\u72c0\u548c\u6bcf\u500b\u8a9e\u97f3\u55ae\u4f4d\u7684\u9577\u5ea6(duration)\uff0c\u5408\u6210 \u5668\u53ef\u900f\u904e\u6539\u8b8a LPC residual \u7684\u9577\u5ea6\u8abf\u6574 F0\uff0c\u548c\u900f\u904e\u589e\u52a0\u6216\u6e1b\u5c11\u5408\u6210\u7684 pitch \u6578\u76ee\u8abf \u6574\u8a9e\u97f3\u7684\u9577\u5ea6\u3002 \u5e74\u4e2d\u958b\u59cb CVC \u7522\u54c1\u7684\u7814\u767c\uff0c\u65bc 2004 \u5e74\u521d\u7522\u54c1\u4e0a\u5e02\u958b\u59cb\u92b7\u552e\uff0c\u7d93\u904e\u591a \u5e74\u4e0d\u65b7\u5730\u6539\u826f\u6280\u8853\uff0c\u8a9e\u97f3\u8fa8\u8b58\u6e96\u78ba\u7387\u548c TTS \u767c\u97f3\u54c1\u8cea\u624d\u9010\u6f38\u70ba\u4f7f\u7528\u8005\u6240\u63a5\u53d7\uff0c\u4e26\u7372\u5f97 \u570b\u5167\u5916\u5404\u5927\u624b\u6a5f\u88fd\u9020\u516c\u53f8\u63a1\u7528\uff0c\u5167\u5efa\u65bc\u5176\u624b\u6a5f\u7522\u54c1\u4e2d\u884c\u92b7\u5168\u7403\u3002\u96d6\u7136\u5982\u6b64\uff0cCVC \u9084\u662f \u6709\u5f88\u5927\u7684\u6539\u9032\u7a7a\u9593\uff0c\u5728\u5168\u7403\u5e02\u5834\u7af6\u722d\u4ecd\u7136\u5341\u5206\u6fc0\u70c8\u7684\u4eca\u5929\uff0c\u6211\u5011\u7d55\u5c0d\u4e0d\u80fd\u6020\u6162\uff0c\u4ee5\u4e0b\u6211 \u5011\u63d0\u51fa\u4e00\u4e9b CVC \u53ef\u4ee5\u6539\u5584\u7684\u5730\u65b9\uff1a", | |
| "content": "<table><tr><td>\u8868\u4e00\u3001CVC \u652f\u63f4\u7684\u8a9e\u8a00\u7248\u672c \u8a9e\u8a00\u7248\u672c \u4e9e\u6fb3\u6d32 \u53f0\u7063\u53e3\u97f3\u4e2d\u6587\u3001\u5927\u9678\u53e3\u97f3\u4e2d\u6587\u3001\u7cb5\u8a9e\u3001\u97d3\u8a9e\u3001\u65e5\u8a9e\u3001\u6cf0\u8a9e\u3001\u571f\u8033\u5176\u8a9e\u3001\u6fb3 \u5730\u5340 \u6d32\u53e3\u97f3\u82f1\u8a9e \u7f8e\u6d32 \u7f8e\u570b\u53e3\u97f3\u82f1\u8a9e\u3001\u4e2d\u5357\u7f8e\u53e3\u97f3\u897f\u73ed\u7259\u8a9e\u3001\u5df4\u897f\u53e3\u97f3\u8461\u8404\u7259\u8a9e \u6b50\u6d32 \u82f1\u570b\u53e3\u97f3\u82f1\u8a9e\u3001\u5fb7\u8a9e\u3001\u6cd5\u8a9e\u3001\u7fa9\u5927\u5229\u8a9e\u3001\u897f\u73ed\u7259\u8a9e\u3001\u8461\u8404\u7259\u8a9e\u3001\u4fc4\u7f85\u65af\u8a9e\u3001 \u8377\u862d\u8a9e\u3001\u4e39\u9ea5\u8a9e\u3001\u6ce2\u862d\u8a9e\u3001\u6377\u514b\u8a9e\u3001\u745e\u5178\u8a9e\u3001\u5e0c\u81d8\u8a9e CVC \u8b93\u4f7f\u7528\u8005\u80fd\u7528\u8a9e\u97f3\u548c\u624b\u6a5f\u5c0d\u8a71\u4e92\u52d5\uff0c\u800c\u9054\u6210\u5728 hand-free\u3001eye-free \u64cd\u63a7\u624b\u6a5f \u7684\u76ee\u7684\u3002CVC \u5177\u5099\u7684\u529f\u80fd\u5305\u62ec\u8a9e\u97f3\u64a5\u865f\u3001\u8a9e\u97f3\u6307\u4ee4\u3001\u8a9e\u97f3\u67e5\u8a62\u806f\u7d61\u4eba\u3001\u8a9e\u97f3\u9ede\u6b4c\u3001\u8a9e \u97f3\u6717\u8b80\u7c21\u8a0a\u3001\u96fb\u5b50\u90f5\u4ef6\u3001\u884c\u4e8b\u66c6\u5167\u5bb9\u7b49\u3002\u5716\u4e00\u70ba\u82f1\u6587\u7248 CVC \u756b\u9762\uff0c\u4ee5\u4e0b\u662f\u900f\u904e CVC \u9032\u884c\u8a9e\u97f3\u64a5\u865f\u7684\u7bc4\u4f8b\uff1a CVC: \u8acb\u8aaa\u6307\u4ee4 User: \u6253\u96fb\u8a71\u7d66\u4f55\u6cf0\u8ed2 CVC: \u6253\u96fb\u8a71\u5230\u4f55\u6cf0\u8ed2\uff0c\u4f4f\u5bb6\u3001\u516c\u53f8\u3001\u624b\u6a5f\u3001\u6216\u53d6\u6d88 User: \u516c\u53f8 CVC: \u516c\u53f8\uff0c\u64a5\u865f\u4e2d\u8acb\u7a0d\u5019 \u53e6\u4e00\u500b\u7bc4\u4f8b\u662f\u4f7f\u7528\u8005\u76f4\u63a5\u8aaa\u300c\u6253\u96fb\u8a71\u7d66\u4f55\u6cf0\u8ed2\u516c\u53f8\u300d \uff1a CVC: \u8acb\u8aaa\u6307\u4ee4 User: \u6253\u96fb\u8a71\u7d66\u4f55\u6cf0\u8ed2\u516c\u53f8 CVC: \u6253\u96fb\u8a71\u5230\u4f55\u6cf0\u8ed2\u516c\u53f8\uff0c\u78ba\u8a8d\u6216\u53d6\u6d88 User: \u78ba\u8a8d CVC: \u8a00\u4e0d\u540c\u800c\u6709\u6240\u5dee\u7570\uff0cRAM \u5927\u7d04\u9700\u8981 500KB\u3002 (\u4e00)\u3001SIVR SIVR \u6a21\u7d44\u53ef\u5206\u6210\u5e7e\u500b\u90e8\u4efd\uff1a\u4e00\u662f\u5f9e\u8072\u97f3\u4e2d\u62bd\u53d6\u7279\u5fb5\u5411\u91cf (feature extraction)\uff0c\u4e8c\u662f \u8fa8\u8b58\u6642\u6240\u4f7f\u7528\u7684\u8a9e\u97f3\u6a21\u578b(acoustic model)\uff0c\u4e09\u662f\u5c07\u8fa8\u8b58\u7684\u8fad\u5f59\u8f49\u63db\u6210\u97f3\u6a19(word-to-phone conversion)\uff0c\u56db\u662f\u5177\u6587\u6cd5\u9650\u5236\u7684\u8a9e\u97f3\u8fa8\u8b58\u641c\u5c0b\u6f14\u7b97\u6cd5(search algorithm)\uff0c\u5982\u5716\u4e09\u3002 1. \u7279\u5fb5\u5411\u91cf \u9032\u884c\u8a9e\u97f3\u8fa8\u8b58\u6642\uff0c\u6703\u5148\u5c07\u9304\u9032\u4f86\u7684\u8072\u97f3\u8f49\u6210\u7279\u5fb5\u5411\u91cf\uff0c\u8072\u97f3\u53ef\u4f86\u81ea\u624b\u6a5f\u4e0a\u7684\u9ea5 \u514b\u98a8\u3001\u6709\u7dda\u8033\u6a5f\u3001\u6216\u85cd\u82bd\u8033\u6a5f\uff0c\u7531\u65bc\u85cd\u82bd\u8033\u6a5f\u53ea\u80fd\u50b3\u8f38 8kHz, 16-bit PCM \u7684\u8072\u97f3\uff0c \u56e0\u6b64\u6211\u5011\u4f7f\u7528\u7684\u8072\u97f3\u8f38\u5165\u70ba 8kHz, 16-bit PCM\u3002\u6bcf\u79d2\u4e2d\u8072\u97f3\u62bd\u53d6\u51fa 100 \u500b\u7279\u5fb5\u5411\u91cf\uff0c \u6bcf\u500b\u7279\u5fb5\u5411\u91cf\u70ba 16 \u500b\u7dad\u5ea6\uff0c\u7531 8 \u7dad\u7684 MFCC (Mel-Frequency Cepstral Coefficients) \u548c 8 \u7dad\u7684 delta MFCC \u6240\u7d44\u6210\u3002MFCC \u7528 CMS(Cepstral Mean Subtraction)\u505a\u901a\u9053\u6548\u61c9 \u88dc\u511f\u3002 \u62bd\u53d6 \u7279\u5fb5\u5411\u91cf \u8a9e\u97f3\u8f38\u5165 \u7279\u5fb5\u5411\u91cf \u5716\u4e09\u3001SIVR \u5404\u7d44\u6210\u90e8\u5206 \u641c\u5c0b \u6f14\u7b97\u6cd5 \u8fa8\u8b58\u7d50\u679c \u8a9e\u97f3\u6a21\u578b \u6587\u5b57\u8f49\u97f3\u6a19 \u6587\u6cd5\u9650\u5236 CVC \u61c9\u7528\u7a0b\u5f0f SIVR \u6a21\u7d44 TTS \u6a21\u7d44 \u8a9e\u97f3\u8f38\u5165 \u8a9e\u97f3\u8f38\u51fa \u5716\u4e8c\u3001CVC \u7cfb\u7d71\u67b6\u69cb Engine level AP level 2. \u8a9e\u97f3\u6a21\u578b \u8a9e\u97f3\u6a21\u578b\u70ba\u50b3\u7d71\u4ee5\u97f3\u7d20\u70ba\u55ae\u4f4d\u7684\u96b1\u85cf\u5f0f\u99ac\u53ef\u592b\u6a21\u578b(Phoneme-based Hidden Markov Model)\uff0c\u6bcf\u500b\u6a21\u578b\u7531 3 \u500b\u7531\u5de6\u5230\u53f3\u7684\u72c0\u614b(state)\u7d44\u6210\uff0c\u6211\u5011\u4f7f\u7528\u4e09\u806f\u97f3\u7d20\u6a21 \u578b(Triphone)\u52a0\u5f37\u5c0d\u9023\u97f3\u7684\u8fa8\u8b58\u3002\u76f8\u8f03\u65bc Triphone \u6578\u91cf\uff0c\u6211\u5011\u6240\u6536\u96c6\u7684\u8a13\u7df4\u8a9e\u6599\u76f8\u5c0d \u4e0d\u8db3\uff0c\u6b64\u6642\u53ef\u7528\u6c7a\u7b56\u6a39(decision tree) [3][4]\u4f86\u6c7a\u5b9a\u54ea\u4e9b\u985e\u4f3c\u7684\u72c0\u614b(state)\u53ef\u5171\u7528\u53c3\u6578 \u548c\u8a13\u7df4\u8a9e\u6599\uff0c\u6211\u5011\u53ef\u8abf\u6574\u5171\u7528\u7684\u7a0b\u5ea6\u4f86\u63a7\u5236\u53c3\u6578\u91cf\uff0c\u4e00\u65b9\u9762\u907f\u514d\u67d0\u4e9b\u53c3\u6578\u7684\u8a13\u7df4\u8a9e \u6599\u4e0d\u8db3\uff0c\u4e00\u65b9\u9762\u53ef\u4f9d\u624b\u6a5f\u7684\u8cc7\u6e90\u8abf\u6574\u51fa\u6700\u4f73\u7684\u6a21\u578b\u5927\u5c0f\u3002 \u8a13\u7df4\u8a9e\u97f3\u6a21\u578b\u4f7f\u7528 forward-backward \u6f14\u7b97\u6cd5 [5]\u3002\u6bcf\u500b\u8a9e\u8a00\u6211\u5011\u8490\u96c6 100 \u5230 800 \u4eba\u4e0d\u7b49\u7684\u8072\u97f3\uff0c\u6bcf\u4eba\u5ff5 200 \u5230 300 \u500b\u53e5\u5b50\uff0c\u5927\u7d04 25 \u5230 30 \u5206\u9418\u7684\u8a9e\u6599\uff0c\u6bcf\u500b\u4eba\u5538\u7684 \u6587\u7a3f\u7686\u4e0d\u540c\uff0c\u53e6\u5916\u518d\u5538 40 \u5230 60 \u500b\u55ae\u8a5e\uff0c\u9019\u4e9b\u55ae\u8a5e\u6db5\u84cb\u8a72\u8a9e\u8a00\u6240\u6709\u7684\u97f3\u6a19\uff0c\u7528\u4f86\u521d \u59cb\u5316\u8a9e\u97f3\u6a21\u578b (boot model)\u3002\u521d\u59cb\u5316\u6a21\u578b\u7684\u6b65\u9a5f\u5982\u4e0b\uff1a\u6211\u5011\u5148\u53d6\u7d04 10%\u7684\u55ae\u8a5e\u8a9e\u6599\uff0c \u7528\u4eba\u5de5\u6a19\u51fa\u6bcf\u500b\u7d44\u6210\u97f3\u6a19\u7684\u908a\u754c\uff0c\u4e26\u8a13\u7df4 CI model (context-independent model)\uff0c\u518d \u7528\u6b64 CI model \u81ea\u52d5\u6a19\u51fa\u6240\u6709\u55ae\u8a5e\u8a9e\u6599\u7684\u97f3\u6a19\u908a\u754c\uff0c\u4e26\u4ee5\u4eba\u5de5\u91cd\u65b0\u6821\u6b63\u8abf\u6574\uff0c\u4e4b\u5f8c \u518d\u62ff\u6821\u6b63\u5f8c\u7684\u8a9e\u6599\u91cd\u65b0\u8a13\u7df4\u4e00\u6b21 CI model\u3002\u6211\u5011\u7684\u7d93\u9a57\u662f\u521d\u59cb\u5316\u6a21\u578b\u7684\u54c1\u8cea\u5c0d\u6700\u5f8c \u8a9e\u97f3\u6a21\u578b\u7684\u597d\u58de\u6709\u5f88\u5927\u7684\u5f71\u97ff\uff0c\u56e0\u6b64\u5728\u9019\u500b\u968e\u6bb5\u6295\u5165\u8f03\u591a\u7684\u4eba\u529b\u505a\u8a9e\u6599\u7684\u8655\u7406\u3002\u8a13 \u7df4\u5b8c\u6210 CI model \u5f8c\uff0c\u518d\u5c07\u5176\u9010\u6b65\u64f4\u5145\u8a13\u7df4\u6210 RCD model (right-context-dependent model)\u548c Triphone model\uff0c\u6700\u5f8c\u518d\u9032\u884c\u5171\u7528\u53c3\u6578\u7684\u8abf\u6574\uff0c\u4f9d\u624b\u6a5f\u5e73\u53f0\u7684\u8a08\u7b97\u8cc7\u6e90\u548c\u5ba2 \u6236\u7684\u9700\u6c42\uff0c\u8abf\u6574\u51fa\u6700\u4f73\u5927\u5c0f\u7684\u8a9e\u97f3\u6a21\u578b\u3002 3. \u6587\u5b57\u8f49\u97f3\u6a19 \u8fa8\u8b58\u6642\u9808\u5c07\u6b32\u8fa8\u8b58\u7684\u6587\u5b57\u8f49\u6210\u97f3\u6a19\uff0c\u518d\u7531\u8a9e\u97f3\u6a21\u578b\u4e2d\u53d6\u51fa\u5c0d\u61c9\u7684 Phoneme HMM\uff0c\u4e32\u63a5\u7d44\u6210\u4ee5\u8a5e\u70ba\u55ae\u4f4d\u6a21\u578b(word model)\uff0c\u8fa8\u8b58\u6642\u5c07\u8f38\u5165\u7684\u8a9e\u97f3\u7279\u5fb5\u5411\u91cf\u548c word model \u6bd4\u5c0d\u8a08\u7b97\u76f8\u5339\u914d\u7684\u6a5f\u7387\u503c\u3002 \u8868\u4e8c\u3001CVC \u5404\u8a9e\u8a00\u7248\u672c\u6240\u4f7f\u7528\u7684\u6587\u5b57\u8f49\u97f3\u6a19\u65b9\u6cd5 \u65b9\u6cd5 \u8a9e\u8a00 \u767c\u97f3\u898f\u5247\u52a0\u4f8b\u5916 \u767c\u97f3\u5c0d\u61c9\u8868 \u7fa9\u5927\u5229\u8a9e\u3001\u897f\u73ed\u7259\u8a9e\u3001\u8461\u8404\u7259\u8a9e\u3001\u6377\u514b\u8a9e\u3001\u571f\u8033\u5176\u8a9e\u3001\u97d3\u8a9e\u3001 \u65e5\u8a9e\u5047\u540d, \u4fc4\u7f85\u65af\u8a9e\u3001\u5e0c\u81d8\u8a9e\u3001\u6cf0\u8a9e \u6587\u5b57\u97f3\u6a19\u5c0d\u61c9\u8868 \u4e2d\u6587\u3001\u65e5\u8a9e\u6f22\u5b57\u3001\u97d3\u8a9e\u6f22\u5b57 \u6c7a\u7b56\u6a39\u6f14\u7b97\u6cd5\u52a0 \u4f8b\u5916\u767c\u97f3\u5c0d\u61c9\u8868 \u82f1\u8a9e\u3001\u6cd5\u8a9e\u3001\u5fb7\u8a9e\u3001\u8377\u862d\u8a9e\u3001\u4e39\u9ea5\u8a9e\u3001\u745e\u5178\u8a9e \u4f9d\u8a9e\u8a00\u7684\u4e0d\u540c\u6211\u5011\u4f7f\u7528\u4e09\u7a2e\u6587\u5b57\u8f49\u97f3\u6a19\u7684\u65b9\u6cd5\uff0c\u7b2c\u4e00\u7a2e\u662f\u6709\u56fa\u5b9a\u767c\u97f3\u898f\u5247\u7684\u8a9e \u8a00\uff0c\u5f9e\u6587\u5b57\u672c\u8eab\u5373\u53ef\u5c0d\u61c9\u51fa\u97f3\u6a19\uff0c\u4f8b\u5982\u897f\u73ed\u7259\u6587\u548c\u7fa9\u5927\u5229\u6587\uff0c\u4f46\u6709\u4e9b\u5916\u4f86\u8a9e\u6703\u51fa\u73fe \u4f8b\u5916\u7684\u72c0\u6cc1\uff0c\u6b64\u6642\u53ef\u5efa\u7acb\u4e00\u500b\u4f8b\u5916\u767c\u97f3\u7684\u5c0d\u61c9\u8868\u4f86\u89e3\u6c7a\u9019\u500b\u554f\u984c\u3002\u7b2c\u4e8c\u7a2e\u662f\u5b8c\u5168\u6c92 \u6709\u767c\u97f3\u898f\u5247\u7684\u8a9e\u8a00\uff0c\u4f8b\u5982\u4e2d\u6587\u6216\u65e5\u6587\u7684\u6f22\u5b57\uff0c\u53ea\u80fd\u7528\u4e00\u500b\u5c0d\u61c9\u8868\u4f86\u5132\u5b58\u6bcf\u500b\u6587\u5b57\u7684 \u767c\u97f3\u3002\u7b2c\u4e09\u7a2e\u662f\u53ef\u5f9e\u6587\u5b57\u4e2d\u5927\u7565\u731c\u51fa\u767c\u97f3\uff0c\u767c\u97f3\u6709\u67d0\u7a2e\u7a0b\u5ea6\u7684\u898f\u5247\u6027\u4f46\u4e0d\u5920\u660e\u78ba\uff0c \u4f8b\u5982\u82f1\u6587\uff0c\u9019\u985e\u8a9e\u8a00\u7684\u767c\u97f3\u96e3\u4ee5\u7528\u5e7e\u689d\u660e\u78ba\u7684\u898f\u5247\u8868\u9054\u51fa\u4f86\uff0c\u6211\u5011\u5247\u4f7f\u7528\u6c7a\u7b56\u6a39\u6f14 \u7b97\u6cd5 [6]\u4f86\u5efa\u7acb\u4e3b\u8981\u767c\u97f3\u898f\u5247\uff0c\u540c\u6a23\u5730\uff0c\u5c0d\u4e00\u4e9b\u4f8b\u5916\u7684\u767c\u97f3\u4e5f\u53ef\u4ee5\u7528\u5c0d\u61c9\u8868\u4f86\u89e3\u6c7a\u3002 \u8868\u4e8c\u662f CVC \u73fe\u5728\u6240\u652f\u63f4\u7684\u8a9e\u8a00\u6240\u4f7f\u7528\u7684\u6587\u5b57\u8f49\u97f3\u6a19\u65b9\u6cd5\u3002 4. \u641c\u5c0b\u6f14\u7b97\u6cd5\u53ca\u6587\u6cd5\u9650\u5236 \u8fa8\u8b58\u4f7f\u7528\u7684\u641c\u5c0b\u6f14\u7b97\u6cd5\u662f Viterbi Search\u3002SIVR \u53ef\u9032\u884c\u7368\u7acb\u8a5e\u5f59\u8fa8\u8b58\uff0c\u6216\u6709\u6587 \u6cd5\u9650\u5236\u7684\u9023\u7e8c\u8a9e\u97f3\u547d\u4ee4\u8fa8\u8b58\uff0c\u4ee5\u8a9e\u97f3\u64a5\u865f\u3001\u8a9e\u97f3\u547d\u4ee4\u800c\u8a00\uff0c\u8f38\u5165\u7684\u9023\u7e8c\u8a9e\u97f3\u547d\u4ee4\u901a \u5e38\u662f\u6709\u898f\u5247\u7684\uff0c\u4f8b\u5982\u300c\u6253\u96fb\u8a71\u7d66\u4f55\u6cf0\u8ed2\u4f4f\u5bb6\u300d\u662f\u7531\u6253\u96fb\u8a71(\u52d5\u4f5c)\u3001\u4f55\u6cf0\u8ed2(\u4eba\u540d)\u3001\u548c \u4f4f \u5bb6 ( \u4f4d \u5740 ) \u6240 \u7d44 \u6210 \uff0c \u300c \u958b \u555f \u8003\u6b64\u6587\u6cd5\u7d50\u69cb\uff0c\u53ea\u5141\u8a31\u7b26\u5408\u6587\u6cd5\u7d50\u69cb\u7684\u72c0\u614b\u8f49\u79fb\uff0c\u5982\u6b64\u53ef\u964d\u4f4e\u9023\u7e8c\u8a9e\u97f3\u8fa8\u8b58\u7684\u641c\u5c0b \u8907\u96dc\u5ea6\uff0c\u964d\u4f4e\u8a08\u7b97\u91cf\uff0c\u540c\u6642\u63d0\u9ad8\u8fa8\u8b58\u7684\u6e96\u78ba\u7387\u3002 \u4e26\u4e0d\u662f\u6240\u6709\u8a9e\u8a00\u7248\u672c\u90fd\u662f\u4f7f\u7528\u5716\u56db\u7684\u6587\u6cd5\u7d50\u69cb\uff0c\u6709\u4e9b\u8a9e\u8a00\u662f\u5f8c\u7f6e\u52d5\u8a5e\u7684\uff0c\u4f8b\u5982 SOV \u8a9e\u5e8f\u7684\u65e5\u6587\u3001\u97d3\u6587\u548c\u571f\u8033\u5176\u6587\u3002\u4ee5\u300c\u6253\u96fb\u8a71\u7d66\u4f55\u6cf0\u8ed2\u4f4f\u5bb6\u300d\u70ba\u4f8b\uff0c\u97d3\u6587\u7684\u5ff5\u6cd5 \u70ba\u300c\ud558\ud0dc\ud5cc\uc758 \uc9d1\uc73c\ub85c \uc804\ud654\uac78\uae30\u300d\uff0c\u5c0d\u61c9\u5230\u4e2d\u6587\u5ff5\u6cd5\u662f\u300c\u4f55\u6cf0\u8ed2(\ud558\ud0dc\ud5cc\uc758)\u4f4f\u5bb6 (\uc9d1\uc73c\ub85c)\u6253\u96fb\u8a71(\uc804\ud654\uac78\uae30)\u300d\u3002\u5c0d\u9019\u985e\u5f8c\u7f6e\u52d5\u8a5e\u7684\u8a9e\u8a00\uff0c\u6211\u5011\u4f7f\u7528\u5982\u5716\u4e94\u6240\u793a \u5728 word model \u4e4b\u9593\u7684\u72c0\u614b\u8f49\u79fb\u6642\uff0cSIVR \u5141\u8a31\u5148\u8f49\u79fb\u81f3\u975c\u97f3\u6a21\u578b(silence model) start \u958b\u555f \u64ad\u653e \u5176\u4ed6\u55ae\u8a5e\u547d\u4ee4 \u4eba\u540d \u61c9\u7528\u7a0b\u5f0f\u540d \u6b4c\u66f2\u540d \u5716\u4e94\u3001\u5f8c\u7f6e\u52d5\u8a5e\u8a9e\u8a00\u7248\u672c CVC \u4f7f\u7528\u7684\u6587\u6cd5 \u6587\u53e5\u5206\u6790 \u6574\u53e5\u6587\u5b57 \u5716\u4e03\u3001TTS \u5404\u7d44\u6210\u90e8\u5206 \u8a9e\u97f3\u5408\u6210\u5668 \u5408\u6210\u8a9e\u97f3 \u97fb\u5f8b\u6a21\u578b \u767c\u97f3\u8a5e\u5178 POS \u8a5e\u5178 CART\u3002\u5e38\u7528\u7684\u7279\u5fb5\u53c3\u6578\u5305\u62ec\uff1a\u8a72\u97f3\u7bc0\u5728\u7247\u8a9e\u5167\u7684\u4f4d\u7f6e\u3001\u5728\u8a5e\u5167\u7684\u4f4d\u7f6e\u3001\u8a72\u97f3\u7bc0\u524d\u5f8c \u7684\u5f71\u97ff\u4e5f\u4e0d\u540c\uff0c\u56e0\u6b64\u6bcf\u7a2e\u8a9e\u8a00\u6703\u4f7f\u7528\u9069\u5408\u8a72\u8a9e\u8a00\u7684\u7279\u5fb5\u53c3\u6578\u3002\u5408\u6210\u6642\u6211\u5011\u5f9e\u7247\u8a9e\u7684 \u97f3\u6a19\u548c POS tag \u4e2d\u53d6\u5f97\u6bcf\u500b\u97f3\u7bc0\u7684\u7279\u5fb5\u53c3\u6578\uff0c\u7531 CART \u4f9d\u64da\u9019\u4e9b\u7279\u5fb5\u9810\u6e2c\u54ea\u500b\u97f3\u7bc0 \u6709\u91cd\u97f3\u3001\u4ee5\u53ca\u5176\u91cd\u97f3\u7a2e\u985e\uff0c\u914d\u5408\u7dda\u6027\u56de\u6b78\u6cd5(linear regression)\u53bb\u8a13\u7df4\u51fa\u9810\u6e2c\u6574\u6bb5\u7247\u8a9e \u7684 F0 \u6a21\u578b\uff0c\u5373\u53ef\u7d44\u5408\u51fa\u7247\u8a9e\u7684 F0 \u8b8a\u5316\uff0c \u8a9e\u97f3\u8fa8\u8b58\u7684\u5be6\u9a57\u5206\u6210\u5169\u7a2e\uff0c\u7b2c\u4e00\u7a2e\u662f\u6a21\u64ec\u5be6\u9a57(simulation)\uff0c\u6211\u5011\u6536\u96c6\u4ee5\u8a72\u8a9e\u8a00\u70ba\u6bcd \u8a9e\u7684\u8a9e\u8005\u7684\u8072\u97f3\u4f5c\u70ba\u6e2c\u8a66\u8a9e\u6599\uff0c\u8a13\u7df4\u8a9e\u6599\u4e2d\u4e0d\u5305\u542b\u6e2c\u8a66\u8005\u7684\u8072\u97f3\uff0c\u6211\u5011\u4ee5\u9019\u4e9b\u6e2c\u8a66\u8a9e\u6599 \u9a57\u8b49\u6211\u5011\u7cfb\u7d71\u7684\u8fa8\u8b58\u7387\uff0c\u4e26\u5728\u958b\u767c\u904e\u7a0b\u4e2d\u505a\u70ba\u6539\u9032\u6280\u8853\u7684\u4f9d\u64da\u3002\u6e2c\u8a66\u8a9e\u6599\u70ba\u4eba\u7684\u540d\u5b57\uff0c \u5305\u542b\u59d3(last name)\u548c\u540d(first name)\uff0c\u6bcf\u7a2e\u8a9e\u8a00\u81f3\u5c11\u6536\u96c6 4 \u5230 6 \u4eba\u7684\u8072\u97f3\uff0c\u7537\u5973\u5404\u534a\uff0c\u6e2c \u8a66\u8005\u9808\u628a\u6bcf\u500b\u6e2c\u8a66\u4eba\u540d\u5538 2 \u6b21\uff0c\u4ee5\u570b\u5167\u5b8f\u9054\u570b\u969b\u96fb\u5b50\u516c\u53f8(HTC)\u6240\u751f\u7522\u7684 Universal \u5e02\u4e2d\u8208\u8def\u4e0a)\u3001\u4ee5\u53ca\u884c\u9032\u5728\u9ad8\u901f\u516c\u8def\u4e0a\u7a97\u6236\u95dc\u9589\u7684\u6c7d\u8eca\u5167\uff0c\u6c7d\u8eca\u5e73\u5747\u6642\u901f\u70ba 90 \u5230 100 \u516c \u64cd\u4f5c\u624b\u6a5f\uff0c\u4fdd\u969c\u81ea\u5df1\u548c\u4ed6\u4eba\u7684\u751f\u547d\u8ca1\u7522\u5b89\u5168\u3002 \u5305\u62ec\u4e0b\u5217 4 \u7a2e\u8a9e\u97f3\u547d\u4ee4\uff1a \u8b93\u5168\u7403\u5927\u90e8\u5206\u8a9e\u8a00\u7684\u4f7f\u7528\u8005\u6240\u63a5\u53d7\uff0c\u7279\u5225\u662f\u8b93\u4f7f\u7528\u8005\u5728\u958b\u8eca\u7684\u6642\u5019\uff0c\u80fd\u66f4\u5b89\u5168\u5730\u7528\u8a9e\u97f3 \u8a2d 20 \u5230 30 \u500b\u61c9\u7528\u7a0b\u5f0f\uff0c\u6bcf\u500b\u6e2c\u8a66\u8005\u5728\u4e00\u7a2e\u6e2c\u8a66\u74b0\u5883\u4e0b\u5538 100 \u53e5\u7684\u9023\u7e8c\u8a9e\u97f3\u6e2c\u8a66\u8a9e\u53e5\uff0c \u7d93\u9a57\uff0c\u548c\u4e00\u4e9b\u8a9e\u8a00\u7684\u8a9e\u97f3\u8fa8\u8b58\u5be6\u9a57\u7d50\u679c\u3002\u4ee5\u624b\u6a5f\u8072\u63a7\u8edf\u9ad4\u800c\u8a00\uff0cCVC \u7684\u8fa8\u8b58\u7387\u5df2\u7d93\u80fd \u6240\u751f\u7522\u7684 PocketPC Phone\uff0c\u624b\u6a5f\u7684\u8a2d\u5b9a\u70ba\u96fb\u8a71\u7c3f\u4e2d\u6709 200 \u500b\u4eba\u540d\uff0c\u624b\u6a5f\u7cfb\u7d71\u5167\u5927\u7d04\u6709\u9810 \u63f4\u591a\u570b\u8a9e\u97f3\u8fa8\u8b58\u8207 TTS \u7684\u624b\u6a5f\u8072\u63a7\u8edf\u9ad4\uff0c\u672c\u8ad6\u6587\u7279\u5225\u63cf\u8ff0\u4e86 CVC \u591a\u570b\u8a9e\u8a00\u7248\u672c\u7684\u958b\u767c \u91cc\uff0c\u6e2c\u8a66\u8eca\u8f1b\u662f 2000 \u5e74\u751f\u7522\u7684 Nissan Sentra 1.6\u3002\u6e2c\u8a66\u624b\u6a5f\u5305\u62ec HTC\u3001ASUS\u3001BenQ \u672c\u8ad6\u6587\u4ecb\u7d39 CVC \u5546\u7528\u5d4c\u5165\u5f0f\u8a9e\u97f3\u547d\u4ee4\u7cfb\u7d71\uff0cCVC \u662f\u570b\u5167\u552f\u4e00\u3001\u4e5f\u662f\u5168\u7403\u5c11\u6578\u80fd\u652f \u82e5\u5e72\u500b\u97f3\u7bc0\u7684\u985e\u5225\u3001\u8a72\u97f3\u7bc0\u6240\u5728\u7684\u8a5e\u7684 POS \u7b49\uff0c\u5728\u4e0d\u540c\u7684\u8a9e\u8a00\u9019\u4e9b\u7279\u5fb5\u53c3\u6578\u5c0d\u97fb\u5f8b \u7684\u6a19\u6e96\uff0c\u76ee\u524d CVC \u6240\u652f\u63f4\u7684\u8a9e\u8a00\u7248\u672c\u4ecd\u6709\u5e7e\u500b\u8a9e\u8a00\u7121\u6cd5\u9054\u5230\u9019\u500b\u6a19\u6e96\uff0c\u7d30\u7bc0\u5247\u4e0d\u5728\u672c \u8ad6\u6587\u4e2d\u8a0e\u8ad6\u3002 \u5230 6 \u540d\u6e2c\u8a66\u8005\uff0c\u7537\u5973\u5404\u534a\u3002\u6e2c\u8a66\u7684\u74b0\u5883\u70ba\u8fa6\u516c\u5ba4\u5167\u3001\u99ac\u8def\u908a(\u5317\u4e8c\u9ad8\u65b0\u5e97\u4ea4\u6d41\u9053\u65c1\u65b0\u5e97 \u56db\u3001\u7d50\u8ad6\u8207\u6539\u826f CVC \u7684\u60f3\u6cd5 \u8a9e\u7684\u6e2c\u8a66\u8005\u4f86\u5be6\u969b\u4f7f\u7528\uff0c\u6211\u5011\u5f9e\u65c1\u89c0\u5bdf\u4f7f\u7528\u72c0\u6cc1\uff0c\u4e26\u8a18\u9304\u6e96\u78ba\u7387\u3002\u6bcf\u500b\u8a9e\u8a00\u6211\u5011\u627e 4 \u91cd\u97f3\u7a2e\u985e\u3001\u4ee5\u53ca\u6bcf\u500b\u8a5e\u7684 POS\uff0c\u5f9e\u4e2d\u53d6\u51fa\u82e5\u5e72\u7684\u7279\u5fb5\u53c3\u6578\uff0c\u5efa\u7acb\u9810\u6e2c\u97f3\u7bc0\u91cd\u97f3\u7684 \u8b39\uff0c\u9a57\u8b49\u65b9\u6cd5\u4e3b\u8981\u662f\u627e\u5e7e\u4f4d\u4ee5\u8a72\u8a9e\u8a00\u70ba\u6bcd\u8a9e\u7684\u807d\u8005\u8a66\u807d\uff0c\u4ee5\u807d\u8005\u4e3b\u89c0\u8a8d\u5b9a\u807d\u5f97\u61c2\u70ba\u5408\u683c \u53e6\u4e00\u7a2e\u662f\u5be6\u5730\u6e2c\u8a66(field test)\uff0cCVC \u6bcf\u7a2e\u8a9e\u8a00\u7248\u672c\u51fa\u8ca8\u524d\uff0c\u6211\u5011\u6703\u627e\u4ee5\u8a72\u8a9e\u8a00\u70ba\u6bcd \u5c0f\u6642\u7684\u8a9e\u6599\uff0c\u7528\u4eba\u5de5\u6a19\u793a\u51fa\u7247\u8a9e\u908a\u754c\u3001\u8a5e\u908a\u754c\u3001\u97f3\u7bc0\u908a\u754c\u3001\u97f3\u6a19\u908a\u754c\u3001\u91cd\u97f3\u4f4d\u7f6e\u3001 POS tag Regression Trees) [8]\u6f14\u7b97\u6cd5\u9810\u6e2c\u7247\u8a9e\u4e2d\u91cd\u97f3\u7684\u4f4d\u7f6e\u3002\u6211\u5011\u627e\u4e00\u4f4d\u8a9e\u8005\u5538\u5927\u7d04 1 \u5230 1.5 \u672c\u8ad6\u6587\u53ea\u6709\u5c0d\u8a9e\u97f3\u8fa8\u8b58\u9032\u884c\u6bd4\u8f03\u79d1\u5b78\u6027\u7684\u5be6\u9a57\u3002\u8a9e\u97f3\u5408\u6210\u7684\u9a57\u8b49\u65b9\u6cd5\u76ee\u524d\u4ecd\u4e0d\u5920\u56b4 \u5f97\u5b9c\u4e14\u7121\u7279\u6b8a\u7684\u901a\u9053\u96dc\u8a0a\uff0c\u90fd\u53ef\u5f97\u5230\u76f8\u4f3c\u7684\u8fa8\u8b58\u7d50\u679c\u3002 end \u4f4f\u5bb6\u3001\u516c\u53f8\u3001\u624b\u6a5f \u6253\u96fb\u8a71 \u8a9e\u97f3\u55ae\u4f4d \u8cc7\u6599\u5eab \u8a5e\u3001\u7247\u8a9e\u3001\u97f3\u6a19\u3001 \u6539\u8b8a\u97f3\u7bc0\u7684 F0 \u5f62\u72c0\uff0c\u5373\u53ef\u5408\u6210\u51fa\u5176\u4ed6\u8072\u8abf\u7684\u97f3\u7bc0\u3002\u9664\u6b64\u4e4b\u5916\uff0c\u6211\u5011\u8b93\u6574\u500b\u7247\u8a9e\u7684 F0 \u7531\u9ad8\u5230\u4f4e\u8b8a\u5316\uff0c\u9047\u5230\u6709\u8072\u8abf\u7684\u97f3\u7bc0\u6642\u5c07\u8a72\u8072\u8abf\u7684 F0 \u5f62\u72c0\u52a0\u5165\u7247\u8a9e\u7684 F0 \u4e2d\uff0c\u7d44\u5408 \u5176\u4ed6\u8a9e\u8a00\u6c92\u6709\u8072\u8abf\uff0c\u4f46\u6709\u91cd\u97f3\u7684\u8a0a\u606f\uff0c\u6211\u5011\u4f7f\u7528 CART (Classification and \u6e2c\u53ef\u80fd\u548c\u6e2c\u8a66\u4eba\u540d\u7684\u9577\u5ea6\u6709\u95dc (\u4e2d\u6587\u4eba\u540d\u70ba 3 \u500b\u97f3\u7bc0\uff0c\u5176\u4ed6\u8a9e\u8a00\u4eba\u540d\u5927\u591a\u9ad8\u65bc 4 \u500b\u97f3\u7bc0) \u3002 \u5be6\u5730\u6e2c\u8a66\u4f7f\u7528\u7684\u5404\u6b3e\u624b\u6a5f\u9304\u97f3\u983b\u7387\u97ff\u61c9\u548c\u901a\u9053\u6548\u61c9\u6216\u8a31\u4e0d\u76e1\u76f8\u540c\uff0c\u4f46\u53ea\u8981\u9304\u97f3\u97f3\u91cf\u63a7\u5236 \u4e09\u3001\u591a\u570b\u8a9e\u8a00\u8a9e\u97f3\u8fa8\u8b58\u5be6\u9a57 \u53f3\uff0c0 dB \u6642\u5e73\u5747\u8fa8\u8b58\u7387\u5247\u4e0b\u964d\u5230\u7d04 76%\uff0c\u6b64\u6642\u5404\u500b\u8a9e\u8a00\u4e5f\u5448\u73fe\u51fa\u8f03\u5927\u7684\u5dee\u7570\uff0c\u6211\u5011\u63a8 dB \u4e4b\u9593\u3002\u99ac\u8def\u908a\u7684\u566a\u97f3\u7a2e\u985e\u8f03\u591a\u4e14\u8f03\u4e0d\u7a69\u5b9a\uff0c\u56e0\u6b64\u8fa8\u8b58\u6548\u679c\u4e5f\u7565\u4f4e\u4e00\u4e9b\uff0c\u7d04\u70ba 88%\u3002 \u56fa\u5b9a\u7684\u898f\u5247\uff0c\u6839\u64da\u97f3\u7bc0\u5728\u7247\u8a9e\u548c\u8a5e\u5167\u4f4d\u7f6e\u7684\u8b8a\u5316\u7d66\u4e88\u4e0d\u540c\u7684\u6b0a\u91cd(weight)\u3002 \u97f3\u7a0b\u5ea6\u52a0\u5f37\uff0c\u8fa8\u8b58\u7387\u96a8\u4e4b\u5448\u73fe\u8f03\u9ad8\u7684\u4e0b\u964d\u8da8\u52e2\uff0c\u5728 S/N \u70ba 5 dB \u6642\u8fa8\u8b58\u7387\u7dad\u6301\u5728 90%\u5de6 \u8fa6\u516c\u5ba4\u74b0\u5883\u4e0b\u6709 95%\u7684\u6b63\u78ba\u7387\uff0c\u6c7d\u8eca\u5167\u7684\u8fa8\u8b58\u7387 91%\u5247\u4ecb\u65bc\u6a21\u64ec\u6e2c\u8a66 S/N \u70ba 10 dB \u548c 5 \u51fa\u6574\u500b\u7247\u8a9e\u7684 F0 \u8b8a\u5316\u3002\u53e6\u4e00\u500b\u97fb\u5f8b\u8a0a\u606f\u662f\u97f3\u7bc0\u7684\u9577\u5ea6\u8b8a\u5316\uff0c\u5728\u4e2d\u6587\u548c\u7cb5\u8a9e\u6211\u5011\u4f7f\u7528 \u4e5f\u6709\u76f8\u7576\u7684\u7a69\u5065\u6027\uff0cS/N \u70ba 10 dB \u4ee5\u5167\u7684\u566a\u97f3\u53ea\u6703\u7a0d\u5fae\u964d\u4f4e\u8fa8\u8b58\u6548\u679c 2%\u5de6\u53f3\u3002\u96a8\u8457\u566a \u96d6\u7136\u548c\u6a21\u64ec\u6e2c\u8a66\u7684\u6307\u4ee4\u8a9e\u6cd5\u7565\u70ba\u4e0d\u540c\uff0c\u5be6\u5730\u6e2c\u8a66\u5f97\u5230\u7684\u7d50\u679c\u548c\u6a21\u64ec\u6e2c\u8a66\u76f8\u8fd1\uff0c \u8a55\u50f9\uff0c\u7d42\u7aef\u4f7f\u7528\u8005\u4e00\u822c\u8a8d\u70ba\u5df2\u7d93\u9054\u5230\u63a5\u8fd1\u771f\u4eba\u767c\u97f3\u7684\u6c34\u6e96\u3002 \u5728\u5b89\u975c\u74b0\u5883\u4e0b\uff0c\u5404\u500b\u8a9e\u8a00\u57fa\u672c\u4e0a\u90fd\u53ef\u9054\u5230 95%\u4ee5\u4e0a\u7684\u6b63\u78ba\u7387\uff0c\u540c\u6642\u6211\u5011\u7684\u8fa8\u8b58\u6838\u5fc3 \u6f22\u5b57\u5178\u7684\u82f1\u6587\u767c\u97f3\u529f\u80fd\u4e2d\uff0c\u65bc 2006 \u5e74\u4e0b\u534a\u5e74\u63a8\u51fa\u7522\u54c1\u300c\u8cfd\u5fae\u96a8\u8eab\u5178\u300d \uff0c\u5f97\u5230\u76f8\u7576\u4e0d\u932f\u7684 \u7684\u6587\u6cd5\u3002 start \u6253\u96fb\u8a71 \u958b\u555f \u64ad\u653e \u5176\u4ed6\u55ae\u8a5e\u547d\u4ee4 \u4eba\u540d \u61c9\u7528\u7a0b\u5f0f\u540d \u6b4c\u66f2\u540d end \u5716\u56db\u3001CVC \u9023\u7e8c\u8a9e\u97f3\u547d\u4ee4\u7684\u6587\u6cd5 \u4f4f\u5bb6\u3001\u516c\u53f8\u3001\u624b\u6a5f \u6216\u5783\u573e\u6a21\u578b(garbage model)\uff0c\u5716\u516d\u63cf\u8ff0\u5305\u542b\u975c\u97f3\u548c\u5783\u573e\u6a21\u578b\u7684\u6253\u96fb\u8a71\u547d\u4ee4\u7684\u6587\u6cd5\uff0c\u5176 \u4e2d S \u4ee3\u8868\u975c\u97f3\u6a21\u578b\uff0cG \u4ee3\u8868\u5783\u573e\u6a21\u578b\u3002\u975c\u97f3\u548c\u5783\u573e\u6a21\u578b\u7686\u662f 1 \u500b\u72c0\u614b\u7684 HMM\uff0c\u4f7f \u7528\u9019\u5169\u500b\u6a21\u578b\u53ef\u4ee5\u8b93\u4f7f\u7528\u8005\u7684\u8f38\u5165\u8a9e\u97f3\u4e2d\u5305\u542b\u4e00\u4e9b\u8d05\u8a5e\u6216\u7a0d\u5fae\u505c\u9813\u4e0d\u8b1b\u8a71\uff0c\u4f8b\u5982\u4f7f \u7528\u8005\u53ef\u4ee5\u8aaa\u300c\u8acb\u5e6b\u6211\u96fb\u8a71\u7d66\u2026\u55ef\u2026\u4f55\u6cf0\u8ed2\u4ed6\u7684\u2026\u4f4f\u5bb6\u7684\u2026\u96fb\u8a71\u300d \uff0c\u5176\u4e2d\u7684\u300c\u8acb\u5e6b\u6211\u300d \u3001 \u300c\u7d66\u2026\u55ef\u2026\u300d \u3001 \u300c\u4ed6\u7684\u2026\u300d \u3001\u548c\u300c\u7684\u2026\u96fb\u8a71\u300d\u7b49\u4e0d\u5c6c\u65bc\u8fa8\u8b58\u8a5e\u5f59\u7684\u8a9e\u97f3\u6709\u6642\u53ef\u4ee5\u88ab S/G \u904e\u6ffe\u6389\u3002\u4e0d\u904e\u6211\u5011\u5be6\u969b\u4f7f\u7528\u7684\u7d93\u9a57\u767c\u73fe\uff0c\u904e\u6ffe\u8d05\u8a5e\u7684\u6548\u679c\u53ea\u6709\u5728\u8fa8\u8b58\u7684\u8a5e\u5f59\u91cf\u5c11\u7684 \u6642\u5019\u6703\u6bd4\u8f03\u597d\uff0c\u5982\u679c\u67d0\u500b\u4eba\u540d\u548c\u300c\u7d66\u2026\u55ef\u2026\u300d\u5ff5\u8d77\u4f86\u7684\u8072\u97f3\u5f88\u985e\u4f3c\uff0c\u5c31\u5f88\u5bb9\u6613\u767c\u751f \u8aa4\u5224\u7684\u60c5\u5f62\u3002 \u4f7f\u7528\u8005\u6709\u6642\u6703\u5fd8\u8a18\u8a9e\u97f3\u6307\u4ee4\u7684\u78ba\u5207\u8aaa\u6cd5\uff0c\u8981\u89e3\u6c7a\u9019\u500b\u554f\u984c\uff0c\u6211\u5011\u53ef\u4ee5\u70ba\u56fa\u5b9a\u7684 \u8a9e\u97f3\u547d\u4ee4\u52a0\u5165\u5176\u4ed6\u5e38\u7528\u7684\u8aaa\u6cd5\uff0c\u4f8b\u5982\u300c\u6253\u96fb\u8a71\u300d\u4e5f\u53ef\u4ee5\u8aaa\u6210\u300c\u64a5\u96fb\u8a71\u300d\u548c\u300c\u6253\u7d66\u300d \uff0c \u300c\u958b\u555f\u300d\u53ef\u4ee5\u8aaa\u6210\u300c\u6253\u958b\u300d\u548c\u300c\u57f7\u884c\u300d \uff0c\u628a\u9019\u4e9b\u4e0d\u540c\u8aaa\u6cd5\u52a0\u5165\u8fa8\u8b58\u7684\u8a5e\u5f59\uff0c\u53ef\u4ee5\u8b93\u4f7f \u7528\u8005\u89ba\u5f97\u7522\u54c1\u66f4\u5bb9\u6613\u4f7f\u7528\u3001\u66f4\u8070\u660e\u3001\u8868\u73fe\u66f4\u7a69\u5b9a\u3002 (\u4e8c)\u3001TTS \u7531\u65bc CVC \u4e3b\u8981\u61c9\u7528\u7684\u5e73\u53f0\u70ba\u624b\u6a5f\uff0c\u76f8\u5c0d\u8a08\u7b97\u91cf\u3001\u8a18\u61b6\u9ad4\u6bd4\u8f03\u5c0f\uff0c\u56e0\u6b64\u6211\u5011\u5fc5\u9808\u9078 \u64c7\u4f7f\u7528\u8f03\u7cbe\u7c21\u7684 TTS \u6280\u8853\u3002\u5728\u4e2d\u6587\u548c\u7cb5\u8a9e\u6211\u5011\u4ee5\u97f3\u7bc0(syllable)\u70ba\u767c\u97f3\u55ae\u4f4d\uff0c\u53ea\u5b58\u653e\u8072\u8abf \u70ba\u4e00\u8072\u7684\u97f3\u7bc0\uff0c\u5176\u4ed6\u8a9e\u8a00\u5247\u4ee5 diphone \u70ba\u55ae\u4f4d\u3002\u6211\u5011\u7684\u8072\u97f3\u8cc7\u6599\u5eab\u5b58\u7684\u662f 16kHz, 16-bit \u7684\u8072\u97f3\uff0c\u900f\u904e\u58d3\u7e2e\uff0c\u4e00\u500b\u8a9e\u8a00\u7684 TTS \u5927\u5c0f\u53ef\u964d\u4f4e\u5230 300KB \u5230 600KB\uff0c\u4f9d\u8a9e\u8a00\u4e0d\u540c\u800c\u6709 \u56b4\u91cd\uff0c\u4e00\u822c\u4f7f\u7528\u8005\u5c0d CVC \u8a9e\u97f3\u5408\u6210\u54c1\u8cea\u7684\u8a55\u50f9\u662f\u300c\u6a5f\u68b0\u97f3\u8f03\u91cd\uff0c\u4e0d\u904e\u9084\u662f\u80fd\u807d\u5f97\u61c2\u3002\u300d \u5728\u97fb\u5f8b\u6a21\u578b\u65b9\u9762\uff0c\u4e2d\u6587\u548c\u7cb5\u8a9e\u90fd\u662f\u5177\u6709\u8072\u8abf(tone)\u7684\u8a9e\u8a00\uff0c\u4e2d\u6587\u5305\u542b\u8f15\u8072\u5171\u6709 \u4e94\u7a2e\u8072\u8abf\uff0c\u7cb5\u8a9e\u5247\u6709\u4e5d\u7a2e\u8072\u8abf\uff0c\u6211\u5011\u7528\u4eba\u5de5\u8a2d\u8a08\u51fa\u5404\u7a2e\u8072\u8abf\u7684 F0 \u5f62\u72c0\uff0c\u900f\u904e\u5408\u6210\u5668 \u679c\uff0c\u4e5f\u662f\u80fd\u5408\u6210\u51fa\u54c1\u8cea\u4e0d\u932f\u7684\u8a9e\u97f3\u3002\u4f8b\u5982\u6211\u5011\u7684\u82f1\u8a9e\u8a9e\u97f3\u5408\u6210\u53ea\u4f54\u7528\u4e0d\u5230 400KB \u7684\u5927 \u5c0f\uff0c\u4f46\u5176\u5408\u6210\u82f1\u6587\u55ae\u8a5e\u548c 4 \u500b\u8a5e\u4ee5\u5167\u7d44\u6210\u7684\u7247\u8a9e\u54c1\u8cea\u5c31\u9084\u4e0d\u932f\uff0c\u6211\u5011\u5c07\u5176\u61c9\u7528\u65bc\u624b\u6a5f\u82f1 \u5e73\u5747 95.63 88.42 91.68 \u6240\u5dee\u7570\u3002\u6211\u5011\u7684\u8a9e\u97f3\u5408\u6210\u65b9\u6cd5\u9808\u5c0d\u8072\u97f3\u505a\u4e0d\u5c11\u7684\u8655\u7406\uff0c\u56e0\u6b64\u5408\u6210\u51fa\u4f86\u7684\u8072\u97f3\u5931\u771f\u4e5f\u6bd4\u8f03 \u6253\u96fb\u8a71 \u4eba\u540d S/G \u5716\u516d\u3001\u5305\u542b\u975c\u97f3\u548c\u5783\u573e\u6a21\u578b\u7684\u6253\u96fb\u8a71\u547d\u4ee4\u7684\u6587\u6cd5 S/G S/G S/G \u4f4f\u5bb6\u3001\u516c\u53f8\u3001\u624b\u6a5f start end TTS \u7684\u67b6\u69cb\u5982\u5716\u4e03\u6240\u793a\uff0c\u6574\u53e5\u6587\u5b57\u8f38\u5165\u5f8c\u5148\u9032\u884c\u6587\u53e5\u5206\u6790\u8655\u7406\uff1a\u5c0d\u6574\u53e5\u6587\u5b57\u9032\u884c\u65b7 \u6211\u5011\u4e5f\u5c07 CART \u61c9\u7528\u5728\u9810\u6e2c\u7247\u8a9e\u4e2d\u6bcf\u500b\u8a9e\u97f3\u55ae\u5143\u7684\u9577\u5ea6\u4e0a\uff0c\u6211\u5011\u7528 VR \u7684\u8a9e\u97f3 \u6a21\u578b\u5c07\u6240\u6709\u8a13\u7df4\u8a9e\u6599\u81ea\u52d5\u5207\u51fa\u6bcf\u500b\u97f3\u6a19\u908a\u754c\uff0c\u5f9e\u4e2d\u53d6\u51fa\u7279\u5fb5\u53c3\u6578\u5efa\u7acb CART\uff0c\u5e38\u7528 \u7684\u53c3\u6578\u548c\u9810\u6e2c\u91cd\u97f3\u7528\u7684 CART \u985e\u4f3c\uff0c\u5305\u62ec\u8a72\u97f3\u6a19\u5728\u7247\u8a9e\u5167\u7684\u4f4d\u7f6e\u3001\u5728\u8a5e\u5167\u7684\u4f4d\u7f6e\u3001 \u5728\u97f3\u7bc0\u5167\u7684\u4f4d\u7f6e\u3001\u8a72\u97f3\u7bc0\u524d\u5f8c\u82e5\u5e72\u500b\u97f3\u6a19\u70ba\u4f55\u3001\u8a72\u97f3\u6a19\u7684\u91cd\u97f3\u7a2e\u985e\u3001\u4ee5\u53ca\u8a72\u97f3\u6a19\u6240 \u5728\u7684\u8a5e\u7684 POS \u7b49\u7b49\u3002 \u9810\u6e2c\u91cd\u97f3\u7684 CART \u9700\u8981\u5927\u91cf\u4eba\u529b\u9032\u884c\u4eba\u5de5\u6a19\u8a3b\uff0c\u76ee\u524d\u7531\u65bc\u4eba\u529b\u4e0d\u8db3\uff0c\u53ea\u6709\u5148\u91dd \u5c0d\u82f1\u6587\u548c\u97d3\u6587\u5efa\u7acb\uff0c\u5176\u4ed6\u975e\u4e2d\u3001\u7cb5\u3001\u82f1\u3001\u97d3\u7684\u8a9e\u8a00\u5247\u662f\u5c07\u97f3\u6a19\u5c0d\u61c9\u6210\u82f1\u6587\u7684\u97f3\u6a19\uff0c \u7528\u82f1\u6587\u7684 CART \u9810\u6e2c\u8a72\u8a9e\u8a00\u7684\u91cd\u97f3\u4f4d\u7f6e\u548c\u5f62\u72c0\uff0c\u4f46\u6bcf\u7a2e\u8a9e\u8a00\u6703\u4f7f\u7528\u4e0d\u540c\u7684\u7247\u8a9e F0 \u8b8a\u5316\u898f\u5247\uff0c\u4e5f\u6709\u8a72\u8a9e\u8a00\u7684\u81ea\u5df1\u7684\u97f3\u6a19\u9577\u5ea6 CART\u3002\u9019\u7a2e\u65b9\u6cd5\u5408\u6210\u7684\u8a9e\u97f3\u807d\u8d77\u4f86\u9084\u53ef \u4ee5\u88ab\u63a5\u53d7\uff0c\u4e0d\u904e\u97fb\u5f8b\u807d\u8d77\u4f86\u602a\u602a\u7684\uff0c\u4ee5\u5fb7\u6587\u70ba\u4f8b\uff0c\u5fb7\u570b\u7684\u5ba2\u6236\u66fe\u63d0\u904e CVC \u7684 TTS \u807d\u8d77\u4f86\u300c\u50cf\u5916\u570b\u4eba\u5728\u8aaa\u5fb7\u6587\u300d \u3002 PocketPC Phone \u9304\u97f3\uff0c\u5728\u5b89\u975c\u7684\u8fa6\u516c\u5ba4\u74b0\u5883\u4e2d\u9304\u88fd\u3002\u6211\u5011\u5be6\u9a57\u7684\u8fa8\u8b58\u8a5e\u5f59\u91cf\u70ba 200 \u500b\u4eba \u540d\uff0c\u7531\u65bc\u624b\u6a5f\u8072\u63a7\u4e3b\u8981\u5728\u958b\u8eca\u6642\u4f7f\u7528\uff0c\u6211\u5011\u628a\u6e2c\u8a66\u8a9e\u6599\u52a0\u5165 AURORA \u6c7d\u8eca\u566a\u97f3\uff0c\u9032\u884c S/N \u70ba 15dB \u5230 0dB \u7684\u566a\u97f3\u5be6\u9a57\u3002\u6709\u6642\u5019\u6211\u5011\u70ba\u4e86\u914d\u5408\u5ba2\u6236\u51fa\u8ca8\u6642\u7a0b\u800c\u8d95\u5de5\uff0c\u4ee5\u81f4\u6709\u4e9b \u8a9e\u8a00\u7684\u5be6\u9a57\u4e26\u4e0d\u5b8c\u6574\uff0c\u5728\u6b64\u6211\u5011\u9078\u64c7\u5217\u51fa 13 \u7a2e\u5be6\u9a57\u505a\u5f97\u8f03\u5b8c\u6574\u7684\u8a9e\u8a00\uff0c\u8fa8\u8b58\u7387\u7d50\u679c\u5217 \u65bc\u8868\u4e09\u3002 \u8868\u4e09\u3001\u5404\u8a9e\u8a00\u65bc AURORA \u6c7d\u8eca\u566a\u97f3\u4e0b\u6a21\u64ec\u5be6\u9a57\u7684\u6e96\u78ba\u7387(%) S/N \u8a9e\u8a00 \u5b89\u975c 15dB 10dB 5dB 0dB \u53f0\u7063\u53e3\u97f3\u4e2d\u6587 98.03 97.04 96.37 93.09 75.33 \u5927\u9678\u53e3\u97f3\u4e2d\u6587 96.62 96.21 95.21 90.33 71.67 \u7cb5\u8a9e 95.36 94.01 93.97 88.01 71.62 1. \u6253\u96fb\u8a71\u7d66 <\u4eba\u540d> 2. \u6253\u96fb\u8a71\u7d66 <\u4eba\u540d> <\u4f4f\u5bb6\u3001\u516c\u53f8\u3001\u624b\u6a5f> 3. \u67e5\u8a62 <\u4eba\u540d> 4. \u958b\u555f <\u61c9\u7528\u7a0b\u5f0f> \u6e2c\u8a66\u7d50\u679c\u5217\u65bc\u8868\u56db\u3002 \u8868\u56db\u3001\u5404\u8a9e\u8a00\u5be6\u5730\u6e2c\u8a66\u7684\u6e96\u78ba\u7387(%) \u5834\u6240 \u8a9e\u8a00 \u8fa6\u516c\u5ba4 \u99ac\u8def\u908a \u884c\u9032\u6c7d\u8eca\u5167 \u53f0\u7063\u53e3\u97f3\u4e2d\u6587 98.6 92.8 93.5 \u5927\u9678\u53e3\u97f3\u4e2d\u6587 96.2 90.4 92.3 \u6211\u5011\u65bc 2002 1. \u6211\u5011\u5728\u88fd\u4f5c\u67d0\u4e9b\u8a9e\u8a00\u7248\u672c\u6642\u906d\u9047\u4e00\u4e9b\u56f0\u96e3\uff0c\u4f8b\u5982\u963f\u62c9\u4f2f\u6587\u56e0\u70ba\u6587\u5316\u7684\u95dc\u4fc2\uff0c\u5f88\u96e3\u627e \u5230\u5973\u6027\u7684\u8a9e\u8005\u4f86\u9304\u97f3\uff0c\u6709\u4e9b\u8a9e\u8a00\u6211\u5011\u6536\u96c6\u7684\u8cc7\u6599\u4e5f\u660e\u986f\u4e0d\u8db3\u3002\u6211\u5011\u672a\u4f86\u5c07\u52a0\u5f37\u9019\u4e9b \u8a9e\u6599\u7684\u6536\u96c6\uff0c\u63d0\u5347\u9019\u4e9b\u8a9e\u8a00\u7684\u8a9e\u97f3\u8fa8\u8b58\u7387\u3002 2. \u6709\u6642\u70ba\u4e86\u914d\u5408\u5ba2\u6236\u51fa\u8ca8\u800c\u8d95\u5de5\uff0c\u4ee5\u81f4\u6709\u4e9b\u8a9e\u8a00\u7684\u5be6\u9a57\u505a\u5f97\u4e0d\u5920\u56b4\u8b39\uff0c\u53ef\u80fd\u9020\u6210\u6d88\u8cbb \u8005\u62b1\u6028\u8fa8\u8b58\u7387\u4e0d\u597d\u7684\u98a8\u96aa\uff0c\u5e36\u7d66\u5ba2\u6236\u56f0\u64fe\u3002\u6211\u5011\u6b63\u7a4d\u6975\u62db\u52df\u66f4\u591a\u5177\u8a9e\u97f3\u76f8\u95dc\u7814\u7a76\u80cc \u666f\u7684\u4eba\u624d\uff0c\u89e3\u6c7a\u76ee\u524d\u4eba\u529b\u4e0d\u8db3\u7684\u554f\u984c\uff0c\u4e26\u70ba\u6bcf\u500b\u8a9e\u8a00\u5efa\u7acb\u66f4\u56b4\u8b39\u66f4\u5b8c\u5584\u7684\u5be6\u9a57\u6d41\u7a0b\u3002 \u53e5(phrasing)POS\uff0c\u4e5f\u7f3a\u4e4f\u6a19\u8a3b\u597d POS tag \u7684\u6587\u5b57\u8cc7\u6599\uff0c\u9019\u985e\u8a9e\u8a00\u6211\u5011 \u4f7f\u7528\u7c21\u55ae\u7684\u898f\u5247\u505a\u65b7\u53e5\uff1a\u5148\u4ee5\u6a19\u9ede\u7b26\u865f\u65b7\u53e5\uff0c\u82e5\u53e5\u5b50\u4ecd\u592a\u9577\u5247\u8a2d\u5b9a\u7247\u8a9e\u97f3\u7bc0\u6578\u7684\u6700 \u5927\u503c\uff0c\u8d85\u904e\u6700\u5927\u503c\u5373\u5f37\u8feb\u65b7\u53e5\uff0c\u4f46\u65b7\u53e5\u7684\u4f4d\u7f6e\u5fc5\u9808\u843d\u5728\u8a5e\u908a\u754c(word boundary)\u4e0a\uff0c \u4e0d\u80fd\u628a\u4e00\u500b\u8a5e\u5206\u958b\u653e\u5728\u5169\u500b\u7247\u8a9e\u4e2d\u3002\u5728\u6b50\u6d32\u8a9e\u7cfb\u8a5e\u548c\u8a5e\u4e4b\u9593\u6703\u7528\u7a7a\u683c\u5206\u958b\uff0c\u6c92\u6709\u65b7 \u8a5e\u7684\u554f\u984c\uff0c\u4f46\u4e9e\u6d32\u8a9e\u8a00\u5982\u4e2d\u6587\u3001\u6cf0\u6587\u5c31\u5fc5\u9808\u5148\u505a\u65b7\u8a5e\uff0c\u627e\u51fa\u8a5e\u908a\u754c\uff0c\u6cf0\u6587\u4e00\u822c\u4e0d\u4f7f \u7528\u6a19\u9ede\u7b26\u865f\uff0c\u65b7\u8a5e\u5728\u6b64\u8655\u66f4\u70ba\u91cd\u8981\u3002\u9019\u7a2e\u7c21\u55ae\u898f\u5247\u65b7\u53e5\u7684\u54c1\u8cea\u4e26\u4e0d\u597d\uff0c\u4e0d\u904e\u5728\u7247\u8a9e \u9593\u7684\u975c\u97f3\u505c\u9813\u4e0d\u8981\u592a\u9577\u7684\u60c5\u6cc1\u4e0b\uff0c\u5408\u6210\u8a9e\u97f3\u807d\u8d77\u4f86\u9084\u4e0d\u81f3\u65bc\u592a\u7a81\u5140\u3002 \u6709\u7455\u75b5\u3001\u97f3\u6a19\u908a\u754c\u548c pitch \u4f4d\u7f6e\u662f\u5426\u5207\u5f97\u6e96\u78ba\u7b49\u3002\u5982\u679c\u80fd\u4ed4\u7d30\u6aa2\u67e5\u8cc7\u6599\u548c\u4eba\u5de5\u5207\u97f3\u7684\u7d50 \u5e73\u5747 96.07 95.51 94.21 90.25 76.85 96.3 88.4 92.8 \u4fc4\u8a9e 2. \u97fb\u5f8b\u6a21\u578b \u7f8e\u570b\u53e3\u97f3\u82f1\u8a9e 98.90 97.90 96.68 92.58 79.40 \u7cb5\u8a9e 94.8 89.7 3. TTS \u7684\u54c1\u8cea\u9084\u6709\u5f88\u5927\u7684\u6539\u9032\u7a7a\u9593\uff0c\u76ee\u524d\u6709\u4e09\u500b\u6539\u9032\u7684\u65b9\u5411\uff0c\u4e00\u662f\u6536\u96c6\u66f4\u591a\u8a9e\u8a00\u7684\u8a13 91.5 3. \u8a9e\u97f3\u5408\u6210\u5668 \u97f3\u8cc7\u6599\u597d\u58de\u3001\u4eba\u5de5\u8655\u7406\u7684\u54c1\u8cea\u6709\u5f88\u5927\u7684\u95dc\u4fc2\uff0c\u5305\u62ec\u9304\u97f3\u8005\u8072\u97f3\u7684\u7279\u6027\u3001\u6240\u9304\u7684\u8072\u97f3\u662f\u5426 \u82f1\u570b\u53e3\u97f3\u82f1\u8a9e 93.88 94.85 94.21 91.45 77.79 \u5fb7\u8a9e 95.17 95.17 93.65 87.81 75.29 \u6cd5\u8a9e 94.83 95.02 94.08 90.25 76.62 \u7fa9\u5927\u5229\u8a9e 95.77 94.15 93.64 91.56 81.73 \u8377\u862d\u8a9e 94.25 93.12 92.62 88.12 74.75 \u65e5\u8a9e 96.55 96.10 92.40 90.40 81.10 \u4fc4\u8a9e 97.15 95.60 93.62 87.07 75.47 \u7f8e\u570b\u53e3\u97f3\u82f1\u8a9e 93.7 85.2 \u7df4\u8cc7\u6599\uff0c\u70ba\u6bcf\u500b\u8a9e\u8a00\u5efa\u7acb\u97fb\u5f8b\u6a21\u578b\uff0c\u4e8c\u662f\u4f7f\u7528\u66f4\u591a\u7684\u8a9e\u97f3\u55ae\u4f4d\u548c unit selection \u7684\u65b9 90.5 \u82f1\u570b\u53e3\u97f3\u82f1\u8a9e 93.2 83.7 88.5 \u5fb7\u8a9e 95.7 86.3 93.8 \u6cd5\u8a9e 96.5 91.4 92.6 \u5df4\u897f\u53e3\u97f3\u8461\u8404\u7259\u8a9e 95.3 87.6 88.7 \u8377\u862d\u8a9e 92.4 84.0 91.3 96.2 88.3 91.2 \u65e5\u8a9e \u6cd5\uff0c\u6e1b\u5c11\u5c0d\u8a9e\u97f3\u8cc7\u6599\u7684\u8abf\u6574\uff0c\u964d\u4f4e\u5408\u6210\u8a9e\u97f3\u5931\u771f\u7684\u7a0b\u5ea6\uff0c\u4e09\u662f\u5efa\u7acb\u56b4\u8b39\u7684\u9a57\u8b49\u65b9\u6cd5 \u548c\u6d41\u7a0b\uff0c\u78ba\u4fdd\u7522\u54c1\u7684\u54c1\u8cea\u3002 4. \u6211\u5011\u767c\u73fe\u5f88\u591a\u4f7f\u7528\u8005\u525b\u958b\u59cb\u5c0d\u7522\u54c1\u4e0d\u719f\u6089\uff0c\u5538\u51fa\u932f\u8aa4\u7684\u8a9e\u97f3\u6307\u4ee4\u800c\u8aa4\u4ee5\u70ba CVC \u807d\u4e0d \u61c2\u4ed6\u5011\u8aaa\u7684\u8a71\uff0c\u5e7e\u6b21\u8fa8\u8b58\u5931\u6557\u5f8c\u7522\u751f\u7684\u632b\u6298\u611f\u8b93\u4ed6\u5011\u4e0d\u9858\u518d\u4f7f\u7528\u3002\u8981\u8b93\u7522\u54c1\u66f4\u5bb9\u6613 \u8a9e\u97f3\u5408\u6210\u5668\u662f\u57fa\u65bc\u7dda\u6027\u9810\u6e2c\u7de8\u78bc(LPC, \u96d6\u7136\u6211\u5011\u9078\u64c7\u4f7f\u7528\u7cbe\u7c21\u7684 TTS \u65b9\u6cd5\uff0c\u4e0d\u904e\u6211\u5011\u7684\u7d93\u9a57\u767c\u73fe\uff0c\u5408\u6210\u8a9e\u97f3\u7684\u54c1\u8cea\u548c\u8a9e \u897f\u73ed\u7259\u8a9e 96.18 95.37 92.83 89.28 78.00 \u5df4\u897f\u53e3\u97f3\u8461\u8404\u7259\u8a9e 96.20 97.15 95.49 93.35 80.29 \u7fa9\u5927\u5229\u8a9e 97.5 92.3 \u4f7f\u7528\uff0c\u6211\u5011\u9808\u63d0\u5347 CVC \u8a9e\u97f3\u8fa8\u8b58\u6a21\u7d44\u7684\u5bb9\u932f\u80fd\u529b\uff0c\u751a\u81f3\u5f9e\u4f7f\u7528\u8005\u4e0d\u5b8c\u6574\u7684\u8a9e\u97f3\u547d\u4ee4 94.0 97.1 89.4 91.2 \u897f\u73ed\u7259\u8a9e \u4e2d\uff0c\u4e86\u89e3\u4f7f\u7528\u8005\u53ef\u80fd\u7684\u610f\u5716\uff0c\u518d\u900f\u904e\u5c0d\u8a71\u7684\u904e\u7a0b\u89e3\u6c7a\u4f7f\u7528\u8005\u4e0d\u6703\u64cd\u4f5c\u7684\u554f\u984c\u3002</td></tr></table>" | |
| } | |
| } | |
| } | |
| } |