铁路与航空驾驶人需频繁使用的标准作业用语，为应用语音信号检测其疲劳状态提供了可能性，而现有的疲劳检测方法在充分性、准确性和普适性等方面还存在一定的局限。为此，本项目拟探讨一种更为有效的应用语音信号检测人体疲劳的新方法。首先，从疲劳对语音的影响机理入手，研究多模型语音疲劳特征的提取方法，从不同角度充分获取语音中所包含的疲劳信息；其次，提出针对不同个体的个性化最优疲劳特征子集的筛选构建与降维处理方法，实现个体差异化语音中疲劳信息的准确描述；最后，建立多分类器混联的疲劳检测模型，提出模型的个体差异自适应调整方案，实现对不同驾驶人疲劳状态的有效识别。项目的创新之处：一是融合多模型语音特征，解决语音中疲劳信息提取不够完备的问题；二是对语音疲劳特征的个性化筛选与降维，解决个体差异性语音中疲劳信息描述不准的问题；三是提出基于多分类器信息融合的自适应检测模型，解决疲劳检测时的不确定性和个体差异性问题。

The standard operation terms frequently used by railway and aviation drivers provide the possibility of using speech signals to detect their fatigue status. However, the existing fatigue detection methods still have some limitations in adequacy, accuracy and universality. For this reason, this project intends to explore a more practical human fatigue detection method by using driver’s speech signal. Firstly, starting from the influence mechanism of fatigue on speech, the extraction methods of multi-model speech fatigue features are studied to fully obtain the fatigue information contained in speech from different angles. Secondly, a series of selecting, construction and dimension reduction methods of individual optimal fatigue feature subsets for different drivers are proposed to achieve the accurate description of the fatigue information in the individual differentiated speech. At last, a multiple classifiers hybrid connected fatigue detection model is established, and then the detection model’s adaptive adjustment scheme for individual differences is put forward, which can effectively recognize different drivers' fatigue status. The innovations of this project are as follows: one is to solve the problem of inadequate extraction of fatigue information by fusing multiple model speech features; the second is to solve the problem of inaccurate description of fatigue information in individual differences’ speech by personalized selecting and dimension reducing of speech fatigue features; the third is to solve the problem of uncertainty and individual differences in fatigue detection by proposing an adaptive detection model based on information fusion of multiple classifiers.