NEXT SPEECH CANDIDATE SCORING DEVICE, METHOD, AND PROGRAM

• Main Authors: HIGASHINAKA RYUICHIRO, MAKINO TOSHIAKI, MATSUO YOSHIHIRO, BESSHO KATSUTO
• Format: Patent
• Language: English; Japanese
• Published: 19.05.2016
• Subjects: ACOUSTICS; CALCULATING; COMPUTING; COUNTING; ELECTRIC DIGITAL DATA PROCESSING; MUSICAL INSTRUMENTS; PHYSICS; SPEECH ANALYSIS OR SYNTHESIS; SPEECH OR AUDIO CODING OR DECODING; SPEECH OR VOICE PROCESSING; SPEECH RECOGNITION

PROBLEM TO BE SOLVED: To smooth interaction between a system and a user by causing the system to repeat a speech suitable for a next speech.SOLUTION: Speech string evaluation data conversion means 21: refers to a concept base 22 with respect to each speech string evaluation data D, with a set of speech string evaluation data D defined as an input, the speech string evaluation D comprising a combination of a speech string A to become a context, a next speech candidate B, and a label C whether or not the next speech candidate B is suited as a next speech of the speech string A; generates a concept vector E of the speech string A; generates a concept vector F of the next speech candidate B; and generates post-conversion speech string evaluation data H comprising a concept vector G formed by combining the concept vector E and the concept vector F, and the label C. Learning means 23 generates a classification model for calculating a score that classifies any concept vector in a same dimension as that of the concept vector G into one value of the label C, from the set of post-conversion speech string evaluation data H.SELECTED DRAWING: Figure 1 【課題】次発話として相応しい発話をシステムが返すことにより、システムとユーザとのインタラクションが円滑になる。【解決手段】発話列評価データ変換手段２１が、文脈となる発話列Ａと、次発話候補Ｂと、発話列Ａの次発話として次発話候補Ｂが相応しいか否かのラベルＣとの組合せからなる発話列評価データＤの集合を入力とし、各発話列評価データＤに対し、概念ベース２２を参照し、発話列Ａの概念ベクトルＥを生成し、次発話候補Ｂの概念ベクトルＦを生成し、概念ベクトルＥと概念ベクトルＦとを結合した概念ベクトルＧとラベルＣとの組合せからなる変換後発話列評価データＨを生成する。学習手段２３が、変換後発話列評価データＨの集合から、概念ベクトルＧと同次元の任意の概念ベクトルが、ラベルＣの一つの値に分類されるスコアを算出するための分類モデルを生成する。【選択図】図１