Applicability and Normative Data for an Arabic Matrix Sentence Test for Speech Recognition in Noise

噪声环境下语音识别阿拉伯语矩阵句测试的适用性和规范数据

阅读:2

Abstract

INTRODUCTION: A matrix sentence test in Modern Standard Arabic (MSA) was developed with a multicenter approach to ensure high acceptability of the speech material. Normative values were obtained at four study sites located in the Persian Gulf region. Test characteristics were compared to other matrix sentence tests. METHODS: Forty-one native Arabic-speaking adults with normal hearing participated in two test development studies, 22 of whom participated in the normative study of the Arabic matrix sentence test. Speech intelligibility scores for individual words and test lists were obtained to optimize word materials and to verify equivalence of test lists, respectively. Normative speech recognition thresholds (SRTs) for adaptive test conduction were also obtained. RESULTS: Optimization of the test material with regard to homogeneous speech intelligibility of the individual words resulted in equivalent test lists with a mean SRT of -7.2 ± 0.2 dB signal-to-noise ratio (SNR) standard deviation (SD) and a slope at SRT of 13.3 ± 1.1%/dB. A mean SRT of -6.7 ± 1.1 dB SNR was obtained for adaptive SRT estimation. In addition, a training effect similar to matrix sentence tests in other languages was found. CONCLUSION: With respect to homogeneity across individual words and test lists, as well as to normal-hearing SRT values and the slope of the discrimination function, the Arabic matrix sentence test achieves reproducible, efficient, and internationally comparable speech recognition data for native Arabic-speaking listeners.

特别声明

1、本页面内容包含部分的内容是基于公开信息的合理引用;引用内容仅为补充信息,不代表本站立场。

2、若认为本页面引用内容涉及侵权,请及时与本站联系,我们将第一时间处理。

3、其他媒体/个人如需使用本页面原创内容,需注明“来源:[生知库]”并获得授权;使用引用内容的,需自行联系原作者获得许可。

4、投稿及合作请联系:info@biocloudy.com。