Weighting of acoustic cues shifts to frication duration in identification of ...Keiichi Yasu
?
K. Yasu, T. Arai, K. Kobayashi and M. Shindo, “Weighting of acoustic cues shifts to frication duration in identification of fricatives/affricates when auditory properties are degraded due to aging,” In Proc. of the Interspeech, 3152?3156, Lyon, 2013.
Weighting of acoustic cues shifts to frication duration in identification of ...Keiichi Yasu
?
K. Yasu, T. Arai, K. Kobayashi and M. Shindo, “Weighting of acoustic cues shifts to frication duration in identification of fricatives/affricates when auditory properties are degraded due to aging,” In Proc. of the Interspeech, 3152?3156, Lyon, 2013.
5. Page, E.B. (1966)
(おそらく)世界初の作文自動採点(中学生(英語母語話者)の作文の採点目的)
? Page, E.B. (1966). The imminence of grading essays by computer, The Phi Delta Kappa international, Vol..
47, No. 5., 238-243.
? Page is widely acknowledged as the father of automated essay scoring, a multi-disciplinary field
exploring computer evaluation and scoring of student writing, particularly essays (Wikipedia).
単語の長さ、段落の数、キーワードの有無などから評定者による評価を予測する。
基本的な考え方は現在と同じ。
「表層的なものしか見ていない」、「コンピュータは内容を理解していない」などの批判は現在の
研究でも当てはまる。
5
6. Bernstein, et. al. (1989)
(おそらく)世界初の発話(発音)自動採点
? Bernstein, J., Weintraub, M., Cohen, M., & Murveit, H. (1989). Automatic evaluation of English spoken
by Japanese students. In Paper FF10, 118th Acoustic Society of America Meeting, November (Abstract in
Journal of the Acoustic Society of America (Suppl. 1), S77).
母語話者の発話から構築した音響モデルとの差(尤度)をもとに判定している。
(解決できない問題であるが)、音響モデルとの差は音響モデル構築に使用した発話に大きく
依存している。
母語話者に似ていることが重要ではなくなってきた。
6
10. Kondo (2010): 読み上げ文の自動採点
The North Wind and the Sun were disputing which was the stronger when a
traveler came along wrapped in a warm cloak. They agreed that the one who first
succeeded in making the traveler take his cloak off should be considered stronger
than the other.
Then the North Wind blew as hard as he could, but the more he blew the more
closely did the traveler fold his cloak around him; and at last the North Wind gave
up the attempt.
Then the Sun shone out warmly, and immediately the traveler took off his cloak.
And so the North Wind was obliged to confess that the Sun was the stronger of
the two.
10
12. 評価基準: CEFR
12
PHONOLOGICAL CONTROL
C2 As C1
C1
Can vary intonation and place sentence stress correctly in order to express finer shades of
meaning.
B2 Has acquired a clear, natural, pronunciation and intonation.
B1
Pronunciation is clearly intelligible even if a foreign accent is sometimes evident and
occasional mispronunciations occur.
A2
Pronunciation is generally clear enough to be understood despite a noticeable foreign accent,
but conversational partners will need to ask for repetition from time to time
A1
Pronunciation of a very limited repertoire of learnt words and phrases can be understood with
some effort by native speakers used to dealing with speakers of his/her language group
(Council of Europe, 2003:117)
評価基準に基づき評定者を訓練し、一般化可能性理
論、多相ラッシュ?モデルを用いて評定者を選んだ
Kondo (2010)。
13. 相関を検証した特長量(抜粋)
13
特徴量 定義
フィラーの数 mm, ehhなどに代表される非語彙の挿入数
無音ポーズの長さ 100ms以上の無音の時間的長さ
Mean length of run 2つのフィラーあるいは無音ポーズの間の音節の数の平均値
言い直し 言い直しで挿入された音節の数
Pruned syllable per second 言い直し、フィラーを除いた音節の数を総発話時間で割ったもの
強勢音節と非強勢音節の比 非強勢音節の長さの平均値を強勢音節の長さの平均で割ったもの
文間のポーズの数 文の境界で置かれたフィラーおよび無音ポーズの数
フレーズ間のポーズの数 フレーズの境界で置かれたフィラーおよび無音ポーズの数
/i/と/?/の識別率 F1およびF2による/i/と/?/の識別率(判別分析による)
/u?/と/?/の識別率 F1およびF2による/u?/と/?/の識別率(判別分析による)
/?/と/?/の識別率 F1およびF2による/?/と/?/の識別率(判別分析による)
15. 採点方法: 近傍法
15
1.5
2
2.5
3
3.5
4
4.5
0.1 0.2 0.3 0.4 0.5 0.6 0.7
Rank C
Rank B
Rank A
Proto C
Proto B
Proto A
0.30.40.50.60.70.80.9
The ratio of weak syllable to the strong one
Syllablepersecond
20. DCTの例
17-2: When you ask John to help you with your homework frankly, how would you say in the
conversation below
A: ( )
B: Sure
3-1: You walk around the park in the morning. How would you respond in the conversation below.
A: What do you do before you leave home in the morning?
B: ( )
20
23. Tutorial English: 授業内容
準備
? ガイドラインに従って授業の準備をする: “Write down six words about you and your life in the circles
below”
授業
? Pair work: Look at the words that your partner wrote down in Preparation. Ask questions to find out
more information about the words and your partner.
? Language
? Practice: Take turns reading and listening activity to the statements below.
A: I woke up late and missed class again.
B: .
? Speaking: Take turns asking questions about topics below.
food, clubs, TV, and friends
? Wrap-up
23
Nakano, M. (2012). Ed. Reach Out Pre-Intermediate Second Edition. Unit 3, Waseda University International.
25. 発話自動採点システムの開発過程
25
The North Wind
and the Sun
were
disputing…
発話誘出タスクの作成
&評価基準の選択
評定者の訓練 発話データの収集
採点と発話の特徴量の関係を調べる
システムの開発
システムの評価
音響モデルの構築
言語モデルの構築
33. 認識率の低い項目
6-06
Describe your hometown
9-16
Describe Miso soup to your host family in USA, referring to its ingredient, cooking method, and
taste.
16-08
Describe your own personality and explain why you think so.
11-07
Choose someone you know well and describe his or her appearance and personality.
34. 中間的な項目(認識率50-70%)
17-03
You cannot help her because you're busy now.
A: Could you help me with my homework?
B: ( ).
18-05
あなたは中国語を勉強することが重要であると考えている。
A: I believe that studying English is very important. What do you think?
B: ( ).