Adapting Automatic Speech Recognition for Accented Air Traffic Control Communications

The speech echo, in particular, is a specific overlapping phenomenon generated by the A TC communication between the sent and received A TCO speech. Another component of noise, radio transmission noise, is the result of A TC being transmitted over High Frequency (HF), V ery High Frequency (VHF) or Ultra High Frequency (UHF) bandwidths that are susceptible to noise caused by static, radio frequency interference, or thermal noise [38]. Auditory information is encoded using amplitude modulation (AM) because it is less susceptible to the capture effect than frequency modulation (FM) [40]. However, AM signals are also less robust against noise and interference, which has been shown to degrade the performance of ASR models [41]. Due to the complex amalgamation of noises, A TC speech often sounds muffled and unintelligible with a low signal-to-noise ratio (SNR). This poses a problem to ASR systems as the noise will either have to be learned by the ASR model or removed through de-noising measures in pre-processing. However, studies have also shown that the use of de-noisers and enhancers could potentially also result in degraded performance in ASR models, making ASR transcription a tricky problem to address [25].