Englert Marina, Lima Livia, Constantini Ana Carolina, Latoszek Ben Barsties V, Maryn Youri, Behlau Mara
Universidade Federal de São Paulo - Unifesp - São Paulo (SP), Brasil.
Centro de Estudos da Voz - CEV - São Paulo (SP), Brasil.
Codas. 2019 Feb 11;31(1):e20180082. doi: 10.1590/2317-1782/20182018082.
This study aimed to verify the best speech material for the AVQI for Brazilian Portuguese language and identify the best validity results between the auditory perceptual judgment (APJ) and the AVQI score on different speech materials.
We recorded voice samples of 50 individuals (dysphonic and vocally healthy) of several continuous speech (cs) variants (i.e., months of the year, numbers 1 to 20, and CAPE-V sentences) and attached the vowel /a/ in each case. The recorded samples were edited to three different durations of cs variants plus vowel: D1-total speech material; D2-customized speech material without voiceless parts; D3-pre-defined cut-off point speech material. These samples were submitted to three voice experts who judged the overall voice quality; and the AVQI analysis. AVQI's precision and concurrent validity were evaluated considering a Gmean threshold of G<0.5 and G<0.68.
The concurrent validity of AVQI and APJ ranged from r = 0.482 to r = 0.634. Numbers presented higher values for all durations. For G<0.5, the best sensitivity and area under the ROC curve was for CAPE-V sentences at D3 (57.8%; 72%). For G<0.68, numbers 1 to 10 had the best diagnostic accuracy and numbers 1 to 20 had the best sensitivity.
Numbers from 1 to 10 had the best correlation results between APJ and AVQI. For G<0.5, sentences had the best diagnostic accuracy; while for G<0.68, numbers had the best diagnostic accuracy. Numbers are commonly used in the Brazilian clinic routine. According to the validity results and daily clinical practice in Brazil, we suggest the use of numbers as cs for concatenated voice samples of voice quality assessments.
本研究旨在验证巴西葡萄牙语听觉语音质量指数（AVQI）的最佳语音材料，并确定不同语音材料在听觉感知判断（APJ）和AVQI评分之间的最佳效度结果。
我们录制了50名个体（嗓音障碍者和嗓音健康者）的语音样本，这些样本包含几种连续语音（cs）变体（即一年中的月份、1到20的数字以及CAPE-V句子），并在每种情况下附加元音/a/。录制的样本被编辑成三种不同时长的cs变体加元音：D1-总语音材料；D2-无清音部分的定制语音材料；D3-预定义截止点语音材料。这些样本提交给三位嗓音专家，他们对整体嗓音质量进行判断，并进行AVQI分析。考虑到G均值阈值G<0.5和G<0.68，评估了AVQI的精密度和同时效度。
AVQI和APJ的同时效度范围为r = 0.482至r = 0.634。在所有时长下，数字呈现出更高的值。对于G<0.5，D3时长下的CAPE-V句子具有最佳敏感性和ROC曲线下面积（57.8%；72%）。对于G<0.68，1至10的数字具有最佳诊断准确性，1至20的数字具有最佳敏感性。
1至10的数字在APJ和AVQI之间具有最佳相关性结果。对于G<0.5，句子具有最佳诊断准确性；而对于G<0.68，数字具有最佳诊断准确性。数字在巴西临床常规中常用。根据巴西的效度结果和日常临床实践，我们建议使用数字作为语音质量评估中串联语音样本的cs。
