Input-independent Attention Weights Are Expressive Enough: A Study of Attention in Self-supervised Audio Transformers

TL;DR

This paper demonstrates that input-independent attention weights in self-supervised audio transformers are sufficiently expressive, enabling reduced computation without sacrificing performance in speech tasks.

Contribution

The study evaluates multiple attention algorithms, categorizes attention weights, and proposes an initialization method that cuts training and inference time by 20%.

Findings

Attention weights can be categorized into four cases.

Input-independent attention weights achieve comparable performance to standard self-attention.

Proposed initialization reduces training and inference time by 20%.

Abstract

In this paper, we seek solutions for reducing the computation complexity of transformer-based models for speech representation learning. We evaluate 10 attention algorithms; then, we pre-train the transformer-based model with those attention algorithms in a self-supervised fashion and treat them as feature extractors on downstream tasks, including phoneme classification and speaker classification. With the assistance of t-SNE, PCA and some observation, the attention weights in self-supervised audio transformers can be categorized into four general cases. Based on these cases and some analyses, we are able to use a specific set of attention weights to initialize the model. Our approach shows comparable performance to the typical self-attention yet requires 20% less time in both training and inference.