Neuroscience Department, Speech Pathology Unit of Otolaryngology Service, Centro Hospitalar do Porto, Porto, Portugal; Speech Therapy Department, Faculty of Health Sciences, Universidade Fernando Pessoa, Porto, Portugal; Biomedical Engineering Department, Faculty of Engineering, Universidade do Porto, Porto, Portugal.
Speech Therapy Department, Faculty of Health Sciences, Universidade Fernando Pessoa, Porto, Portugal.
J Voice. 2014 Mar;28(2):210-5. doi: 10.1016/j.jvoice.2013.08.001. Epub 2013 Sep 17.
OBJECTIVES/HYPOTHESIS: The aim of this article was to describe the results of an audio-perceptual evaluation carried out by 10 judges, on a database comprising 90 voice recordings plus 10 samples repetition, with the purpose of characterizing the intra- and interrater reliability.
Exploratory, transversal.
The classification of the GRBAS parameters was obtained for each one of the 10 experts, concerning the 90 voice samples. The intraclass correlation coefficient determined the interrater reliability. For the 10 repeated voices, the intrarater reliability was assessed by means of a dispersion analysis.
The average judges' classification for each of the GRBAS parameters differs (P < 0.05). The values of the correlations, with confidence intervals of 95%, between the average scores for all components of the GRBAS scale lie, in general, between 0.838 and 0.966. The first three parameters of the scale (G, R, and B) have the higher interrater reliability. Differences were statistically significant (P < 0.05) for experts 1, 6, 9, and 10, which means a poor intrarater reliability for 40% of the judges.
All the experts had similar evaluation criteria for the assessment of the five parameters of the GRBAS scale (the values of the confidence intervals at 95% of the experts average ratings of the GRB were above 0.8). However, its quantification is not statistically similar. Asthenia and Strain have lower reliability. Most experts do not reveal statistically significant differences between the values assigned to the GRB parameters (P > 0.05).
目的/假设：本文的目的是描述由 10 名评委进行的音频感知评估的结果，该评估基于包含 90 个语音录音和 10 个重复样本的数据库，旨在描述组内和组间可靠性。
探索性、横断面研究。
对 10 名专家中的每一位进行了 GRBAS 参数的分类，涉及 90 个语音样本。组内相关系数确定了组间可靠性。对于 10 个重复的声音，通过分散分析评估了组内可靠性。
每位评委对每个 GRBAS 参数的平均分类结果不同（P<0.05）。在 GRBAS 量表所有成分的平均分数之间，置信区间为 95%的相关值通常在 0.838 到 0.966 之间。该量表的前三个参数（G、R 和 B）具有更高的组间可靠性。对于专家 1、6、9 和 10，差异具有统计学意义（P<0.05），这意味着 40%的评委的组内可靠性较差。
所有专家对 GRBAS 量表五个参数的评估标准相似（专家平均评分的 95%置信区间内的 GRB 值大于 0.8）。然而，其量化结果并不具有统计学意义上的相似性。乏力和紧张度的可靠性较低。大多数专家对 GRB 参数分配的值没有显示出统计学上的显著差异（P>0.05）。
