The stress state of pilots is a core human factor affecting aviation safety, which has attracted significant attention in the field of aviation safety in recent years. However, the accurate monitoring of this state remains a key challenge for the practical application of related technologies. To address this issue, this paper proposes a pilot stress detection model based on a CNN-BiLSTM hybrid neural network, designed for aviation human factor analysis tasks. First, 160-dimensional spectral statistical features of audio are extracted using MFCC (Mel-Frequency Cepstral Coefficients). Then, CNN (Convolutional Neural Network) is employed to efficiently extract local discriminative features, while BiLSTM (Bidirectional Long Short-Term Memory) captures the long-term contextual dependencies of feature sequences. The synergy between these two components enhances the accuracy of recognition. Experimental results show that the detection model proposed in this study exhibits excellent classification performance, achieving an accuracy of 96% and a weighted F1-score of 94%. Comparative experiments with single CNN and BiLSTM models verify the effectiveness of the component design of the proposed model. Meanwhile, combined with confusion matrix analysis, it is further proved that the model has good classification ability for imbalanced data. In addition, the lightweight design of the model reduces computational costs, enabling its deployment on airborne embedded devices and providing a reliable technical solution for the real-time stress monitoring of pilots.