Multi-space channel representation learning for mono-to-binaural conversion based audio deepfake detection

Rui Liu, Jinhua Zhang, Guanglai Gao

Abstract: Highlights•We pioneer the definition of a feature representation of binaural audio as three spaces.•We present MSCR-ADD, a novel ADD scheme with a novel multi-space channel representation learning strategy.•Experimental results on four datasets verify the power of our method.