Speech Quality Evaluation of Air Traffic Control Based on Phoneme Level

In the real air traffic control scene, the control scene is complex and changeable. In the actual control process, we often encounter pilots with different accents and timbre, and the amount of professional control speech data with labels is small. In this paper, the phoneme-level-based pronunciation quality evaluation method is used to map the divided individual phonemes to the corresponding audio frames, and the likelihood score values obtained from the actual phonemes and the target text phonemes are compared with the likelihood score values obtained without knowing the corresponding text, and the likelihood ratio is used as a good or bad evaluation of speech pronunciation quality. Compared with the traditional method of using the distance between real pronunciation and standard pronunciation to evaluate pronunciation quality, the method proposed in this paper is more generalized, not affected by the sparse or fragmented speech data of the evaluation content because it is more difficult to obtain, and fits the real situation of complex and changeable air traffic control scenarios. By accurately detecting pronunciation errors and providing evaluation feedback, the accuracy of communication between pilots and controllers can be guaranteed, and the level of air traffic safety can be guaranteed. At the same time, the speech quality evaluation of air traffic control is also an important basis for the intelligent and intelligent development of civil aviation air traffic control, promoting the development of speech technology and service ability of air traffic control, and steadily promoting the high-quality and safe development of civil aviation in the new era. The statistics of the dichotomous prediction results were used to represent the performance of the pronunciation quality evaluation model, where the rejection recognition rate (FRR) was 10.57%.