認知症者とロボットの対話のための相槌認識における話者依存性の分析と話者適応の効果について  [in Japanese] On an analysis of speaker dependency and speaker adaptation for the recognition of back-channeling expressions in dialog between an robot and persons with dementia  [in Japanese]

Search this Article

Author(s)

Abstract

軽度認知症者への情報支援を目的として,対話を通じて必要な情報を伝達するために,ロボットの問いかけに対する相槌を11の機能カテゴリに識別する実験を行った.対話のタスクドメインを限定した言語モデルと,高齢者用音響モデルを用いた実験により,平均およそ6割の識別率を得た.また,話者毎の分析から,識別率は,認知症の評価スケールとの相関は低いが,年齢との間には強い負の相関が確認された.特に,80歳を超える高齢者の識別率を改善するため話者適応を行ったところ,識別率は平均7割まで向上した.さらに,肯定・納得とそれ以外のカテゴリを識別する2値分類においては,平均8割の識別率を得た.

For the purpose of establishing a dialog between a robot and persons with dementia to remind information, this paper concerns classification of back-channeling expressions into 11 functional categories. An off-line experiment shows that the classification rate of 60% in average is yielded by a classifier using a language model for a fixed task domain and an acoustic model for elderly persons. It is also shown that the classification rate correlates negatively with the age and not with a scale of dementia. To improve the classification rate for persons older than 80 years, speaker adaptation is shown to be effective and increase the rate up to 70%. For binary classification into affirmative expression and the others, the classification rate is further improved up to 80%.

Journal

  • IEICE technical report. Welfare Information technology

    IEICE technical report. Welfare Information technology 111(226), 61-65, 2011-09-29

    The Institute of Electronics, Information and Communication Engineers

References:  15

Codes

  • NII Article ID (NAID)
    110008899564
  • NII NACSIS-CAT ID (NCID)
    AA1155230X
  • Text Lang
    JPN
  • Article Type
    ART
  • ISSN
    0913-5685
  • NDL Article ID
    023340741
  • NDL Call No.
    Z16-940
  • Data Source
    CJP  NDL  NII-ELS 
Page Top