Real World Neural Speech Enhancement, Separation, and Dereverberation for Monaural, Multichannel, and Multi-modal Setup
ECE 590SIP, February 8, 2023, Zhongweiyang Xu
Abstract:
The presentation will mainly cover the speech enhancement, separation, and dereverberation problems in general. Different applications like real-time communication, hearing aids, speech augmented reality, speech codec, and their corresponding algorithm requirements will be mentioned. Current state of single-channel speech enhancement models and multichannel methods like WPE, masked-MVDR will be covered in detail. At last, some important open problems like array-agnostic setup, visual speech enhancement, and my own related work will be mentioned in brief.