E-005
Step-by-Step Speech Enhancement Using Stacked Hourglass Wave Networks

◎Aliaksei Khadanovich・Shuichi Arai（東京都市大）

We propose a novel method for speech enhancement, which is based on Stacked Hourglass Networks by Park Sungheon et al.. The proposed method operates directly in the time domain and it enables to train to reduce different types of noises step-by-step rather than at once. This idea was achieved only through a stacked feature of the baseline model, which consists in connecting several encoder-decoder modules consistently. The experimental results on Voice Bank corpus (VCTK) dataset evaluation show that our method achieves competitive results compared to the other methods in the speech enhancement task.