E-Branchformer: Branchformer with Enhanced Merging for Speech Recognition
Authors:
Top Cited Papers
- 9 January 2023
- conference paper
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 84-91
Abstract
Conformer, combining convolution and self-attention sequentially to capture both local and global information, has shown remarkable performance and is currently regarded as the state-of-the-art for automatic speech recognition (ASR). Several other studies have explored integrating convolution and self-attention but they have not managed to match Conformer's performance. The recently introduced Branchformer achieves comparable performance to Conformer by using dedicated branches of convolution and self-attention and merging local and global context from each branch. In this paper, we propose E-Branchformer, which enhances Branchformer by applying an effective merging method and stacking additional point-wise modules. E-Branchformer sets new state-of-the-art word error rates (WERs) 1.81% and 3.65% on LibriSpeech test-clean and test-other sets without using any external training data.Keywords
All Related Versions
This publication has 29 references indexed in Scilit:
- Attention-Based ASR with Lightweight and Dynamic Convolutions2020
- A Comparative Study on Transformer vs RNN in Speech Applications2019
- Attention Based On-Device Streaming Speech Recognition with Large Speech Corpus2019
- Jasper: An End-to-End Convolutional Neural Acoustic Model2019
- Improved Training of End-to-end Attention Models for Speech Recognition2018
- ESPnet: End-to-End Speech Processing Toolkit2018
- Non-local Neural Networks2018
- Self-Attention with Relative Position Representations2018
- Identity Mappings in Deep Residual Networks2016
- Connectionist temporal classification2006