Obstructive sleep apnea (OSA) is a sleep disorder in which pharyngeal collapse during sleep causes complete (apnea) or partial (hypopnea) airway obstruction. OSA is common and can have severe implications, but often remains undiagnosed. The most widely used objective measure of OSA severity is the number of obstructive events per hour of sleep, known as the apnea-hypopnea index (AHI). This study reports an innovative system to identify OSA subjects while they are awake, not asleep, using speech signal processing techniques. The assumption is that OSA affects the acoustic parameters of speech because it is associated with anatomical and functional abnormalities of the upper airway. The system associates three different sub-systems based on features extracted from breathing segments within continuous speech signals, information acquired from sustained vowels using a convolutional neural network, and inherent information in continuous speech signals using a recurrent neural network. Each of these sub-systems provided an AHI estimation and were combined with age and body mass index (BMI) to produce a composite system that estimates AHI using a linear regression. The sample was composed of 398 subjects (men and women). The performance of each sub-system was examined separately, in addition to the composite system. As expected, the composite AHI estimation yielded the superior results, with a Pearson correlation coefficient of 0.61 between the estimated and diagnosed AHI. To distinguish between OSA and non-OSA subjects, a classification decision was made using an AHI threshold of 15 events per hour. The system achieved an average accuracy of 77.14%, a sensitivity of 75%, and a specificity of 79%.

中文翻译：

---

使用来自清醒受试者的语音信号诊断阻塞性睡眠呼吸暂停

阻塞性睡眠呼吸暂停 (OSA) 是一种睡眠障碍，其中睡眠期间咽部塌陷导致完全（呼吸暂停）或部分（呼吸不足）气道阻塞。OSA 很常见，可能会产生严重的影响，但往往未被确诊。最广泛使用的 OSA 严重程度的客观衡量标准是每小时睡眠中阻塞性事件的数量，称为呼吸暂停低通气指数 (AHI)。该研究报告了一种创新系统，可在 OSA 对象醒着而不是睡着时使用语音信号处理技术识别他们。假设 OSA 会影响语音的声学参数，因为它与上呼吸道的解剖和功能异常有关。该系统根据从连续语音信号中的呼吸段提取的特征关联三个不同的子系统，使用卷积神经网络从持续元音中获取的信息，以及使用循环神经网络从连续语音信号中获取的固有信息。这些子系统中的每一个都提供了 AHI 估计值，并与年龄和体重指数 (BMI) 相结合，生成了一个使用线性回归估计 AHI 的复合系统。样本由 398 名受试者（男性和女性）组成。除了复合系统之外，还分别检查了每个子系统的性能。正如预期的那样，复合 AHI 估计产生了优异的结果，估计和诊断 AHI 之间的 Pearson 相关系数为 0.61。为了区分 OSA 和非 OSA 受试者，使用每小时 15 个事件的 AHI 阈值做出分类决定。该系统平均准确率为 77.14%，