This study investigated acoustic features and prosodic strategies used by both healthy speakers and speakers with dysarthria, to produce a perceptually-detectable sentence accent. Accordingly, 80 adult speakers (50 with a speech impairment) were asked to produce 3 pairs of sentences with different accent positions. All speech samples were perceptually judged by three experts, and were acoustically analyzed. The fundamental frequency, intensity, and duration were not only analyzed within the syllable, but also in contrast with the previous syllable, and in contrast with the entire sentence. These features were used as input for a linear discriminant analysis. This newly-developed acoustic approach reveals that healthy speakers mainly rely on the following features to produce a perceptually-detectable accent: a change in frequency within the target syllable, with a simultaneous increase of intensity and contrast in the frequency between the target syllable and the previous syllable. Speakers with dysarthria mainly use the contrast in frequency and intensity between the target syllable and the previous syllable, rather than the contrasting with the rest of the sentence. They also use durational parameters as an element in prosodic accent production. Although both groups use some common features, they differ significantly (p < 0.01) in the way they realize accent production. The results of this study show that sentence accent production in dysarthric speech can be adequately described by a set of acoustic features. Until now, such description in the literature was primarily based on perceptual evaluations. The current approach may assist in objective assessments and more appropriate therapy methods.

这项研究调查了健康说话者和患有构音障碍的说话者所使用的声学特征和韵律策略，以产生可觉察到的句子重音。因此，要求80位成人说话者（其中50位有语言障碍）产生3对重音位置不同的句子。所有语音样本均由三位专家进行感知判断，并进行了声学分析。基本频率，强度和持续时间不仅在音节中进行了分析，而且还与以前的音节以及整个句子进行了对比。这些特征用作线性判别分析的输入。这种新开发的声学方法表明，健康的说话者主要依靠以下功能来产生可感知的口音：目标音节内的频率变化，同时强度和目标音节与前一个音节之间的频率对比度增加。具有构音障碍的说话者主要使用目标音节和前一个音节之间的频率和强度上的对比，而不是与句子其余部分的对比。他们还使用持续时间参数作为韵律重音产生的要素。尽管这两组人都有一些共同的特征，但他们实现口音产生的方式却有很大差异（p <0.01）。这项研究的结果表明，通过一组声学特征可以充分地描述构音障碍语音中句子重音的产生。到目前为止，文献中的此类描述主要基于感知评估。