An online model for vowel imitation learning

详细信息    查看全文

• 作者：Heikki Rasilo^a ; ^b ; ^{heikki.rasilo@aalto.fi} ; Okko Rä ; sä ; nen^a ; ^{okko.rasanen@aalto.fi}
• 关键词：Imitation ; Correspondence problem ; Normalization problem ; Vowel learning ; Speech acquisition ; Associative algorithm ; Weakly supervised learning
• 刊名：Speech Communication
• 出版年：2017
• 出版时间：February 2017
• 年：2017
• 卷：86
• 期：Complete
• 页码：1-23
• 全文大小：2880 K
• 卷排序：86

文摘

A virtual infant (LeVI) learns to imitate Finnish vowels in interaction with humans. LeVI associates its babbles to imitative responses online with weakly-supervised learning. Initially LeVI does not know the number or characteristics of native vowels. Initially LeVI does not know the acoustic consequences of its articulations. Learning occurs online despite ambiguity in the responses and inaccuracy in babbling.