Text-to-Speech Application for Training of Aviation Radio Telephony Communication Operators

Air traffic control (ATC) and its dedicated radio telephony communication are critical components of safe and efficient air traffic. After the COVID-19 pandemic, the aviation industry faced a shortage of air traffic controllers (ATCos) and pilots, highlighting a significant problem: managing resources for training new ATCos and pilots. This article explores using a text-to-speech application (TTS app) to simulate aviation radio telephony communication. The app utilizes open-source pretrained TTS models fine-tuned using publicly available ATC communication-specific datasets. It synthesizes textual ATC utterances to simulate ATCo instructions and pilot responses, creating a realistic two-way communication scenario. It includes twenty ATCos and eight pilot voice models developed using two distinct fine-tuning approaches: 1) An end-to-end TTS method leveraging deep learning techniques, and 2) a voice cloning method supporting multilingual speech generation. The app was evaluated in an online study by 20 international study subjects, comprising 14 ATCos, four pilots, and two individuals from other aviation backgrounds. The performance of the voice models varied across different aspects of audio quality such as overall experience, clarity, pronunciation, intonation, naturalness, and speed due to more than 4100 subjective rating values. The voice cloning models were rated significantly better overall than the end-to-end models. The female voice cloning models were rated significantly better overall than the male voice cloning models—both fine-tuned with ATCo data. The majority of voice cloning models especially for ATCo utterances received average overall ratings between 4 and 4.5 out of the highest score of 5. More than 83% of ratings classified the audio articulation speed as optimal. While some issues on pauses, pronunciation, and volume consistency were noted, the general feedback, especially on realism, showed the feasibility of TTS for ATC communication training. The possibility to synthesize speech faster than real-time and initial explorations of large language models for TTS show that developing operational downstream applications is on the horizon.