Improving air traffic control speech intelligibility by reducing speaking rate effectively

摘要

Low intelligibility of Air Traffic Control (ATC) speech is one major cause of aircraft accidents every year. Many factors can affect speech intelligibility, among which the most prominent aspects is the high speaking rate commonly present in ATC speech. Hence, a possible solution would be to improve intelligibility by artificially lengthening the spoken utterance to lower the speaking rate. In this work, we explore the lengthening of clean recorded ATC utterances by first identifying phoneme sequences in a given utterance. Such identified phoneme segments can then be lengthened. We will examine effects of lengthening vowels-only, consonants-only, or homogeneous lengthening. To verify our approach, we will conduct human listening test to evaluate the intelligibility. The results show 74.67% was obtained in AB preference test.