PARALANGUAGE INFORMATION ESTIMATION MODEL LEARNING DEVICE, PARALANGUAGE INFORMATION ESTIMATION DEVICE AND PROGRAM

To highly precisely estimate paralanguage information even if speech which is difficult to specify the paralanguage information is used for model learning.SOLUTION: An acoustic feature extraction section 11 extracts an acoustic feature from speech. A reverse teacher determination section 12 determin...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: KOBASHIGAWA SATORU, KAMIYAMA HOSONA, ANDO ATSUSHI
Format: Patent
Sprache:eng ; jpn
Schlagworte:
Online-Zugang:Volltext bestellen
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:To highly precisely estimate paralanguage information even if speech which is difficult to specify the paralanguage information is used for model learning.SOLUTION: An acoustic feature extraction section 11 extracts an acoustic feature from speech. A reverse teacher determination section 12 determines a reverse teacher label representing a reverse teacher which is not a correct answer of the paralanguage information of the speech on the basis of a paralanguage information label representing a determination result of the paralanguage information which a plurality of listeners give at every speech. A reverse teacher estimation model learning section 13 learns a reverse teacher estimation model outputting posterior probability of the reverse teacher to the input acoustic feature on the basis of the acoustic feature extracted from the speech and the reverse teacher label.SELECTED DRAWING: Figure 2 【課題】パラ言語情報を特定することが困難である発話をモデル学習に用いた場合であっても、高精度にパラ言語情報を推定する。【解決手段】音響特徴抽出部11は、発話から音響特徴を抽出する。逆教師決定部12は、複数の聴取者が発話ごとに付与したパラ言語情報の判定結果を表すパラ言語情報ラベルに基づいて、その発話のパラ言語情報の正解ではない逆教師を表す逆教師ラベルを決定する。逆教師推定モデル学習部13は、発話から抽出した音響特徴と逆教師ラベルとに基づいて、入力された音響特徴に対する逆教師の事後確率を出力する逆教師推定モデルを学習する。【選択図】図2