Hieda I, Kuchinomachi Y
National Institute of Bioscience and Human Technology, Ibaraki, Japan.
Percept Mot Skills. 1997 Dec;85(3 Pt 2):1483-91. doi: 10.2466/pms.1997.85.3f.1483.
To improve the naturalness of synthesized voices, the relations between the physical characteristics of the synthesized voices and the psychological effects should be established. The authors performed a psychological evaluation using natural voices of men and women as stimuli. The method of principal component analysis was applied to intercorrelations, the numerical ratings of the evaluation, and principal components were extracted which represented aspects ordinary people use to evaluate natural voices. Pitches of the voices used in the evaluation were analyzed as samples of physical voice parameters, and the relations between the pitches and the principal components were examined. Four principal components were extracted, representing aspects to which most people were observed to pay most attention when listening to voices. A significant relation was also found between physical pitches which were standardized by sex and the perceived pitches which were introduced from the principal component scores. This finding suggests that different criteria are used for perceptions of pitches of men and women.
为提高合成语音的自然度，应建立合成语音的物理特征与心理效应之间的关系。作者以男性和女性的自然语音作为刺激进行了心理评估。主成分分析法应用于相互关系、评估的数值评分，并提取了代表普通人用于评估自然语音的各个方面的主成分。评估中使用的语音音高作为语音物理参数的样本进行了分析，并研究了音高与主成分之间的关系。提取了四个主成分，代表大多数人在听语音时最关注的方面。在按性别标准化的物理音高与从主成分得分引入的感知音高之间也发现了显著关系。这一发现表明，男性和女性对音高的感知使用了不同的标准。
