Abstract: As one of the most representative applications built on deep learning, audio systems, including keyword spotting, automatic speech recognition, and speaker identification, have recently been ...