Application of Speech Recognition Technology in Air-Ground Communication Based on Bidirectional Recurrent Neural Network and Transfer Learning

With the rapid development of the aviation industry, ensuring the accuracy of air-ground communication and improving the real-time efficiency of conversations is crucial for flight safety. This paper proposes a speech recognition technology solution based on Bidirectional Recurrent Neural Networks (Bi-RNN) combined with Transfer Learning, aimed at further optimizing and enhancing speech recognition performance in air-ground communication. By analyzing the unique requirements of China's civil aviation air-ground communication, the Bi-RNN model is applied to process speech signals, and the Connectionist Temporal Classification (CTC) algorithm is used for model training. On this basis, Transfer Learning methods are employed to further optimize the model, improving its accuracy in complex speech environments. The study results indicate that, compared to traditional Deep Neural Network (DNN) models, the proposed algorithm significantly enhances speech recognition accuracy, with a reduction in error rate by 2.05%. The paper also analyzes the challenges faced by this technology in practical applications. These technical achievements provide support for further advancing the intelligence level of aviation communication and offer a solid theoretical foundation.