韩语中声学粗糙度指数的交叉验证

文献信息

中文摘要

目的： 声学粗糙度指数（ARI）是一种多变量声学测量方法，专门用于通过整合多种声学特征（包括传统声学测量难以捕捉的次谐波现象）来客观量化嗓音粗糙度。由于ARI分析的是包含连续语音的拼接语音样本，其性能可能受到语言特定语音和韵律特征的影响。尽管其临床效用已在日本和德国人群中得到证实，但尚未在韩语使用者中得到验证。因此，本研究旨在验证ARI，并建立区分韩语使用者中发声障碍嗓音与正常嗓音的最佳诊断截断值。
方法： 分析了330名韩语母语者（74名对照组和256名发声障碍患者）的语音录音。使用由持续元音和朗读段落组成的拼接语音样本计算ARI评分。三名经验丰富的言语语言病理学家使用7点听觉感知量表独立评估嗓音粗糙度，平均评分作为参考标准。使用Spearman等级相关评估同时效度，并进行受试者工作特征（ROC）曲线分析以确定最佳诊断截断值。
结果： ARI与听觉感知粗糙度评分呈强正相关（Spearman ρ = 0.804，P < 0.001）。发声障碍组的ARI评分显著高于对照组（P < 0.001）。ROC分析显示出极好的诊断准确性（AUC = 0.923），最佳截断值为2.29，灵敏度为74.6%，特异度为90.5%。
结论： ARI在区分韩语使用者中发声障碍嗓音与正常对照方面表现出强同时效度和极好的诊断性能。提出的截断值2.29提供了临床有用的区分能力，并支持ARI作为听觉感知嗓音评估的客观辅助工具的跨语言适用性。

英文摘要

OBJECTIVES: The Acoustic Roughness Index (ARI) is a multivariate acoustic measure specifically developed to objectively quantify vocal roughness by integrating multiple acoustic features, including subharmonic phenomena that are difficult to capture using conventional acoustic measures. Because the ARI analyzes concatenated voice samples that include connected speech, its performance may be influenced by language-specific phonetic and prosodic characteristics. Although its clinical utility has been demonstrated in Japanese and German populations, it has not been validated in Korean speakers. Therefore, this study aimed to validate the ARI and establish an optimal diagnostic cutoff for discriminating dysphonic voices from normal voices in Korean speakers.
METHODS: Voice recordings from 330 native Korean speakers (74 controls and 256 patients with dysphonia) were analyzed. Concatenated voice samples consisting of a sustained vowel and a reading passage were used to calculate ARI scores. Three experienced speech-language pathologists independently rated vocal roughness using a 7-point auditory-perceptual scale, and the mean rating served as the reference standard. Concurrent validity was evaluated using Spearman's rank correlation, and receiver operating characteristic (ROC) curve analysis was performed to determine the optimal diagnostic cutoff.
RESULTS: ARI demonstrated a strong positive correlation with auditory-perceptual roughness ratings (Spearman's ρ = 0.804, P < 0.001). ARI scores were significantly higher in the dysphonia group than in the control group (P < 0.001). ROC analysis demonstrated excellent diagnostic accuracy (AUC = 0.923), and the optimal cutoff value was 2.29, yielding a sensitivity of 74.6% and a specificity of 90.5%.
CONCLUSIONS: The ARI demonstrated strong concurrent validity and excellent diagnostic performance for discriminating dysphonic voices from normal controls in Korean speakers. The proposed cutoff of 2.29 provides clinically useful discrimination and supports the cross-linguistic applicability of the ARI as an objective adjunct to auditory-perceptual voice assessment.