Eadie Tanya L, Doyle Philip C
Department of Speech and Hearing Sciences, University of Washington, Seattle, WA, USA.
J Voice. 2005 Mar;19(1):1-14. doi: 10.1016/j.jvoice.2004.02.002.
The purpose of this study was (1) to determine the relationship between acoustic measures and auditory-perceptual dimensions of overall voice severity and pleasantness and (2) to evaluate the ability of acoustic and auditory-perceptual measures to discriminate normal from dysphonic voices. Thirty adult dysphonic speakers and six, age-matched normal control speakers were asked to provide oral reading samples of the Rainbow Passage. Acoustic analysis of the speech samples was used to identify abnormal phonatory events associated with dysphonia. The acoustic program calculated long-term average spectral measures, glottal noise measures, and those measures based on linear prediction (LP) modeling. Twelve adult listeners judged overall voice severity and pleasantness from the connected speech samples using direct magnitude estimation (DME) procedures. The acoustic measures accounted for 48% of overall voice severity and 40% of voice pleasantness for dysphonic speakers. The classification performance of the acoustic measures and auditory-perceptual measures was quantified using logistic regression analysis. When acoustic measures or auditory-perceptual measures were considered in isolation, classification was generally accurate and similar across measures. Classification accuracy improved to 100% when acoustic and auditory-perceptual measures were combined. These data provide further support for use of both auditory-perceptual evaluation and acoustic analyses for classifying and evaluating dysphonia.
（1）确定声学指标与整体嗓音严重程度和愉悦度的听觉感知维度之间的关系；（2）评估声学和听觉感知指标区分正常嗓音与发声障碍嗓音的能力。30名成年发声障碍者和6名年龄匹配的正常对照者被要求提供《彩虹段落》的口头朗读样本。对语音样本进行声学分析，以识别与发声障碍相关的异常发声事件。声学程序计算长期平均频谱指标、声门噪声指标以及基于线性预测（LP）建模的指标。12名成年听众使用直接量值估计（DME）程序从连续语音样本中判断整体嗓音严重程度和愉悦度。对于发声障碍者，声学指标占整体嗓音严重程度的48%，占嗓音愉悦度的40%。使用逻辑回归分析对声学指标和听觉感知指标的分类性能进行量化。当单独考虑声学指标或听觉感知指标时，分类通常是准确的，且各指标之间相似。当声学指标和听觉感知指标结合使用时，分类准确率提高到100%。这些数据为使用听觉感知评估和声学分析来分类和评估发声障碍提供了进一步的支持。
