Input-independent Attention Weights Are Expressive Enough: A Study of Attention in Self-supervised Audio Transformers
This work addresses efficiency improvements for speech processing models, though it appears incremental as it builds on existing attention mechanisms.
The paper tackles the problem of reducing computational complexity in transformer-based speech representation learning by evaluating 10 attention algorithms and categorizing attention weights into four general cases. The result is an approach using specific attention weights for initialization that achieves comparable performance to typical self-attention while requiring 20% less time in training and inference.
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.