Awan Shaheen N, Roy Nelson
Department of Audiology and Speech Pathology, Bloomsburg University of Pennsylvania, Bloomsburg, PA 17815-1301, USA.
J Speech Lang Hear Res. 2009 Apr;52(2):482-99. doi: 10.1044/1092-4388(2009/08-0034).
The purpose of this experiment was to assess the ability of an acoustic model composed of both time-based and spectral-based measures to track change following voice disorder treatment and to serve as a possible treatment outcomes measure.
A weighted, four-factor acoustic algorithm consisting of shimmer, pitch sigma, the ratio of low-to-high frequency spectral energy, and a measure of the cepstral peak was used to predict dysphonia severity in pre- and post-treatment vowel samples from 88 women with primary muscle tension dysphonia treated by manual circumlaryngeal therapy. Predicted severity ratings were also compared to mean perceived severity ratings determined by a group of judges.
Predicted severity scores were strongly associated with perceived dysphonia severity ratings for pretreatment, posttreatment, and change in dysphonia severity. Analyses of the agreement between predicted and perceptual severity ratings indicated that the majority of differences were within +/- 1 standard deviation from the mean difference. Acoustic predictions of perceived severity were observed to be most accurate for the midportion of the 7-point equal-appearing interval severity scale.
The acoustic model and predicted dysphonia severity scores show promise as a sensitive and objective outcomes measure, even with extremely perturbed pre-treatment voice samples that would be difficult to analyze using traditional time-based perturbation measures.
本实验的目的是评估一个由基于时间和基于频谱的测量组成的声学模型跟踪嗓音障碍治疗后变化的能力，并作为一种可能的治疗结果测量方法。
使用一种加权的四因素声学算法，该算法由声门波颤动、基频标准差、低频与高频频谱能量之比以及谐波峰值测量组成，来预测88名接受手法环甲膜治疗的原发性肌肉紧张性发声障碍女性治疗前和治疗后元音样本中的发声障碍严重程度。预测的严重程度评分还与一组评判员确定的平均感知严重程度评分进行了比较。
预测的严重程度得分与治疗前、治疗后以及发声障碍严重程度变化的感知发声障碍严重程度评分密切相关。对预测和感知严重程度评分之间的一致性分析表明，大多数差异在平均差异的±1个标准差范围内。观察到在7点等距严重程度量表的中间部分，感知严重程度的声学预测最为准确。
即使对于使用传统基于时间的微扰测量方法难以分析的极度紊乱的治疗前嗓音样本，声学模型和预测的发声障碍严重程度评分也有望作为一种敏感且客观的结果测量方法。
