arXiv 2005.08100
Conformer: Convolution-augmented Transformer for Speech Recognition
By Anmol Gulati, James Qin, et al.
Published 2020-05-16
Citation lineage
Review the prior work and downstream research connected to this paper.
Recently Transformer and Convolution neural network (CNN) based models have shown promising results in Automatic Speech Recognition (ASR), outperforming Recurrent neural networks (RNNs). Transformer models are good at capturing content-based global interactions, while CNNs exploit local features effectively. In this work, we achieve the best of both worlds by studying how to combine convolution neural networks and t…