arXiv 2005.08100
Conformer: Convolution-augmented Transformer for Speech Recognition
By Anmol Gulati, James Qin, et al.
Published 2020-05-16
Mindmap
Browse the paper's core ideas, clusters, and relationships in a structured outline.
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…