Hazan Valerie, Markham Duncan
Department of Phonetics and Linguistics, University College London, London WC1E 6BT, United Kingdom.
J Acoust Soc Am. 2004 Nov;116(5):3108-18. doi: 10.1121/1.1806826.
This study investigated acoustic-phonetic correlates of intelligibility for adult and child talkers, and whether the relative intelligibility of different talkers was dependent on listener characteristics. In experiment 1, word intelligibility was measured for 45 talkers (18 women, 15 men, 6 boys, 6 girls) from a homogeneous accent group. The material consisted of 124 words familiar to 7-year-olds that adequately covered all frequent consonant confusions; stimuli were presented to 135 adult and child listeners in low-level background noise. Seven-to-eight-year-old listeners made significantly more errors than 12-year-olds or adults, but the relative intelligibility of individual talkers was highly consistent across groups. In experiment 2, listener ratings on a number of voice dimensions were obtained for the adults talkers identified in experiment 1 as having the highest and lowest intelligibility. Intelligibility was significantly correlated with subjective dimensions reflecting articulation, voice dynamics, and general quality. Finally, in experiment 3, measures of fundamental frequency, long-term average spectrum, word duration, consonant-vowel intensity ratio, and vowel space size were obtained for all talkers. Overall, word intelligibility was significantly correlated with the total energy in the 1- to 3-kHz region and word duration; these measures predicted 61% of the variability in intelligibility. The fact that the relative intelligibility of individual talkers was remarkably consistent across listener age groups suggests that the acoustic-phonetic characteristics of a talker's utterance are the primary factor in determining talker intelligibility. Although some acoustic-phonetic correlates of intelligibility were identified, variability in the profiles of the "best" talkers suggests that high intelligibility can be achieved through a combination of different acoustic-phonetic characteristics.
本研究调查了成人和儿童说话者可懂度的声学语音相关因素,以及不同说话者的相对可懂度是否取决于听众特征。在实验1中,对来自同一口音组的45名说话者(18名女性、15名男性、6名男孩、6名女孩)的单词可懂度进行了测量。材料由124个7岁儿童熟悉的单词组成,这些单词充分涵盖了所有常见的辅音混淆;刺激音在低水平背景噪声中呈现给135名成人和儿童听众。7至8岁的听众比12岁的听众或成人犯的错误明显更多,但个体说话者的相对可懂度在各群体中高度一致。在实验2中,对实验1中确定为可懂度最高和最低的成人说话者,获取了他们在多个语音维度上的听众评分。可懂度与反映发音、语音动态和总体质量的主观维度显著相关。最后,在实验3中,获取了所有说话者的基频、长期平均频谱、单词时长、辅音-元音强度比和元音空间大小的测量值。总体而言,单词可懂度与1至3千赫兹区域的总能量和单词时长显著相关;这些测量值预测了可懂度中61%的变异性。个体说话者的相对可懂度在不同听众年龄组中显著一致,这一事实表明,说话者话语的声学语音特征是决定说话者可懂度的主要因素。尽管确定了一些可懂度的声学语音相关因素,但“最佳”说话者的特征存在变异性,这表明通过不同声学语音特征的组合可以实现高可懂度。