15 Development of an online infrastructure for teaching Japanese prosody based on information processing of speech and text corpora (Private Edition) Nobuaki MINEMATSU (Professor, Graduate School of Engineering, The University of Tokyo) This paper describes how speech and text corpora were used in developing OJAD (Online Japanese Accent Dictionary), an online Japanese prosody teaching/learning system. Current problems related to teaching Japanese prosody are summarized, and the relationship between these problems and system development is explained. A corpus of spoken verbs along with their conjugations was used to build a module to conduct verb accent search. A text corpus of sentences with both accentual phrase boundaries and accent nuclei labeled was used to train a boundary detector and an accent nucleus detector. These detectors were used to construct a prosodic reading tutor. Subjective assessment was done by 80 teachers of Japanese to both verb accent search module and prosodic reading tutor. All the teachers assessed the search module as very effective or effective to some degree. The reading tutor was evaluated by 73 teachers as very effective or effective to some degree. These results indicate high effectiveness of the two systems. Through the development of OJAD, the author has become aware that there are still gaps to be bridged in communication between Japanese teachers and speech engineers pertaining to the needs of the former and the technology being made available by the latter. These gaps are pointed out in the future directions. Online Japanese Accent Dictionary OJAD
1 2013 400 Jenkins 2009 Pinet 2010 public speaking 2014 2009 2013, OJAD 2015a 16 12 public speaking Online Japanese Accent Dictionary (OJAD) 2012 8 2013, OJAD 2015a 2015 4 11 2014 11 2015 4 16 2012 11 2015 4 63 OJAD OJAD 2015b OJAD 2013 OJAD
2 2.1 2009 2010 2014 OJAD NHK NHK 1998 High/Low High(H)/Low(L) H/L H/L
1 H/L H/L 2.2 2009 1
2 2009 3 2014, 2014 2009 1 1) 2) 3) 2009 2.3 2.1 2.2 1) 2) 3) 4) 5) 2 3 2009 CD OJAD INTERSPEECH2013 OJAD 2015c URL 15
JEITA (Japan Electronics and Information Technology Industries Association) IT-4006 JEITA2010 2 4 JEITA 3 3.1 12 16 3,500 42,000 2013, 2014 web 42,000 12 / 200msec 4 JEITA
2015 3.2 web MySQL v5.1.63 web CakePHP v2.1.3 3 1) 2) 3) 3) 4) ON/OFF 18 19 4 PC 4 4 MP3 3.3 80 2/3 2013 2 80 1 4 4.1 1983
にほんごのべんきょうは / むずかしいですが / だいすきです あ はアクセント核 図 1: フレージングとポージングに基づく韻律指導 2006 年の調査によると 日本全国で 約 33%の家庭がペットを買っているそうです ニセ ン/ロク ネンノ/チョ ーサニヨルト ニホンゼ ンコクデ ヤ ク%/サ ンジュー/ サンパーセ ントノ/カテーガ/ペ ットオカッ. テイルソ ーデス%. グッズの専門店もでき お洒落な服を着た犬も よく見かけます グ ッズノ/センモ ンテンモ/デ キ オシャ レナ/フ%ク オ/キ%タイヌモ ヨ ク/ミカケマ ス%. :アクセント核 /:アクセント句境界 :ポーズ %:母音の無声化 図 2: 漢字仮名混じり文から JEITA フォーマットへの変換例 図 3: 検索 表示条件 図 4: 単語検索結果の例
5 2013 2014 CRF (Conditional Random Fields) 2007 Japanese News Article Sentences, JNAS 2015 6 1 1 7 / 6,334 6,109 2013 2014 CRF(CRF++v0.57) F 94.1% F 96.7% TASET (Tokyo Accent Sandhi Estimation Toolkit) 2013 4.2 3 2.2 2009 2009 5
/ L H H L Minematsu and Hirose 1995 3 5 A B (A,B)=(, ) (, ) (, ) (, ) A H L L B L A B H B H L L A, B B H 6 1971 4.3 4.1 4.2 / 1) 2) 3) 4) JEITA 2 OJAD KDDI N2 KDDI 2015 2015 4 N2
1: % 21.7 4.5 11.1 4.5 9.6 3.5 8.6 2.0 8.1 1.0 7.6 0.5 7.6 0.5 5.6 3.5 a) 71.0 29.0 0.0 0.0 2: % b) 38.7 59.7 1.6 日 本 の 漫 画 は 面 白 いし,アニメも 大 好 きです 形 態 素 解 析 アクセント 句 境 界 推 定 アクセント 核 位 置 推 定 にほんの まんがは おもしろいし あにめも だいすきです アクセント 句 接 続 規 則 LHHH LHHH LHHHLL HLLL HLLLLL ( 上 級 者 用 モーラ 別 H/L 値 ) LHHH HHHH HHHHLL HLLL HLLLLL ( 初 級 者 用 モーラ 別 H/L 値 ) 基 本 周 波 数 パターン 生 成 過 程 モデル 基 本 周 波 数 パターン 生 成 過 程 モデル 上 級 者 用 ピッチパターン 初 級 者 用 ピッチパターン 5: 6:
JEITA 6 Fast, Normal, Slow 4.4 3.3 80 2013 3 OJAD Public speaking web OJAD web 3 3.3 8.5% 1) 2) 5 OJAD 5.1 OJAD 2012 8
2012 8 Google Analytics 2015 4 38 16 7 11 10 4 2 5.2 OJAD 2012 11 3 4 OJAD 2015 4 63 OJAD 2015b OJAD OJAD 2.3 OJAD 1983 30 CALL OJAD
OJAD OJAD OJAD 6 2014 OJAD 6 OJAD (Online Japanese Accent Dictionary) OJAD OJAD Project OJAD OJAD 2015a facebook OJAD OJAD 2015a [ 2010] (2010) SP2009-151 19 24. [ 2014] (2014) [NHK 1998] NHK(1998) NHK NHK. [ 2014] (2014) Project OJAD http://youtu.be/aqv8xsxp7dq http://youtu.be/ijcibditq_g (2015 4 29 ) [ 2015] (2015) Japanese News Article Sentences (JNAS), http://research.nii.ac.jp/src/jnas.html (2015 4 29 ) [KDDI 2015] KDDI (2015) N2 TTS DSK http://www.kddilabs.jp/products/audio/n2tts/product.html (2015 4 29 ) 6
[ 2007] (2007) SP2006-174 31 36. [ 1983] (1983) J66-D(7) 849 856. [ 2013] (2013) J96-D(3) 644 654. [ 2013] (2013) Tokyo Accent Sandhi Estimation Toolkit (TASET), https://sites.google.com/site/suzukimasayuki/accent (2015 4 29 ) [ 2015] (2015) (A) http://jisho.jpn.org (2015 4 29 ) [JEITA2010] (2010) JEITA IT-4006 http://www.jeita-speech.org (2015 4 29 ) [ 2009] (2009) 18 45 51. [ 2009], (2009). [ 2013] (2013) http://www.jpf.go.jp/j/about/press/dl/0927.pdf (2015 4 29 ) [ 2014] (2014) CRF 1-R5-28 443 444. [ 2009] (2009) 65(2) 69 80. [ 2014] (2014), 7 45 71. [ 1971] (1971) 27(9) 445 453. [OJAD 2015a] Project OJAD(2015) Online Japanese Accent Dictionary, http://www.gavo.t.u-tokyo.ac.jp/ojad/ (2015 4 29 ) [OJAD 2015b] Project OJAD(2015) OJAD http://www.gavo.t.u-tokyo.ac.jp/ojad/pages/workshop (2015 4 29 ) [OJAD 2015c] Project OJAD(2015) OJAD http://youtu.be/kpjifu2abxg (2015 4 29 ) [ 2013] (2013) J96-D(10) 2496 2508.
[Jenkins 2009] Jenkins, J. (2009) World Englishes: a resource book for students, Routledge. [Minematsu and Hirose 1995] N. Minematsu and K. Hirose(1995) Role of prosodic features in the human process of perceiving spoken words and sentences in Japanese, J. Acoust. Soc. Japan(E), 16(5), 311 320. [Pinet 2010] Pinet, M., P. Iverson and M. Huckvale(2010) Second-language experience and speechin-noise recognition: the role of L2 experience in the talker-listener accent interaction, Proc. SLaTE (CD-ROM).
3: % a) b) 62.7 42.6 28.8 50.0 8.5 7.4 0.0 7: 2012 8 2015 4