1 paper · 1 filter
Jinjiang Liu, Xueliang Zhang
For dual-channel speech enhancement, it is a promising idea to design an end-to-end model based on the traditional array signal processing guideline and the manifold space of multi…