1. Phase-Aware Deep Speech Enhancement: It’s All About The Frame Length [Paper]
This paper is submitted to interspeech 2022. focuses on the role of phase and amplitude in DNN-based speech enhancement at different STFT frame lengths. It is concluded that DNN-based phase estimation benefits from shorter frames. On the contrary, the best performance is achieved with 32ms frames when the only amplitude is processed.