Visual speech recognition using compact hypercomplex neural networks

Iason-Ioannis Panagos, Giorgos Sfikas, Christophoros Nikou

Abstract: Highlights•Significant model compression with Parameterized Hypercomplex Layers.•Minor accuracy trade-off depending on the compression level.•Negligible accuracy loss for the audio recognition models.•Better performance than some much larger models.•Potential for applications in low resource devices due to small size.