Stochastic Attention Head Removal: A simple and effective method for improving Transformer Based ASR Models

TL;DR

This paper introduces a simple stochastic head removal technique during training that enhances Transformer-based ASR models by reducing redundancy and improving performance across multiple datasets.

Contribution

The paper proposes a novel stochastic attention head removal method during training, leading to more efficient and better-performing Transformer and Conformer ASR models.

Findings

Consistent performance improvements on WSJ, AISHELL, Switchboard, and AMI datasets.

Achieved state-of-the-art results on Switchboard and AMI datasets.

Reduces redundancy in attention heads, enhancing model efficiency.

Abstract

Recently, Transformer based models have shown competitive automatic speech recognition (ASR) performance. One key factor in the success of these models is the multi-head attention mechanism. However, for trained models, we have previously observed that many attention matrices are close to diagonal, indicating the redundancy of the corresponding attention heads. We have also found that some architectures with reduced numbers of attention heads have better performance. Since the search for the best structure is time prohibitive, we propose to randomly remove attention heads during training and keep all attention heads at test time, thus the final model is an ensemble of models with different architectures. The proposed method also forces each head independently learn the most useful patterns. We apply the proposed method to train Transformer based and Convolution-augmented Transformer (Conformer) based ASR models. Our method gives consistent performance gains over strong baselines on the Wall Street Journal, AISHELL, Switchboard and AMI datasets. To the best of our knowledge, we have achieved state-of-the-art end-to-end Transformer based model performance on Switchboard and AMI.