基于语谱图的管制员疲劳状态检测研究

在现阶段利用陆空通话语音对管制员的疲劳状态的研究中，大多只考虑了语音在时域或频域的变化，而忽视了疲劳会同时在时域与频域上产生影响。本文将三种疲劳状态下的陆空通话语音分别转化为可同时反应时域与频域特性的语音频谱图像，利用灰度共生矩阵（GLCM）提取四维典型的特征参数，对比管制员在不同状态下特征参数的变化情况，确认所选特征具有较好的区分度，将所选特征作为管制员疲劳检测模型的输入特征进行检测。结果表明：利用语谱图特征结合传统特征作为输入特征的检测准确率最高，达到95.49%，较单一使用传统特征的检测准确率高出4%；管制员疲劳状态的变化会直观地反映在语谱图上，并会对其特征值产生影响，利用这种影响对管制员疲劳状态进行检测，可以得到良好的检测结果。

Research On Controller Fatigue State Detection Based On Speech Spectrogram

In the current research on the fatigue state of controllers using radiotelephony communication, most of them only consider the changes of voice in the time domain or frequency domain, while ignore that fatigue will affect the time domain and frequency domain at the same time. In this paper, the voice of radiotelephony communication in the three fatigue states is converted into speech spectrum images that can reflect the characteristics of both the time domain and frequency domain, and the grayscale co-existence matrix (GLCM) is used to extract the typical feature parameters in four dimensions, compare the changes of the characteristic parameters of the controllers in different states, and confirm that the selected features have a good discrimination. The selected features were detected as the input features of the controller fatigue detection model, and the detection accuracy of using the spectral pattern features combined with the traditional features as the input features was the highest, reaching 95.49%, which was 4% higher than that of the traditional features alone. The results show that the change of controllers fatigue state will be intuitively reflected on the spectrogram and will have an impact on its eigenvalues, and good results can be obtained by using this influence to detect the controllers fatigue state.