Stabilising and accelerating light gated recurrent units for automatic speech recognition

TL;DR

This paper introduces SLi-GRU, a stabilized and faster training version of Light Gated Recurrent Units, improving training stability and speed while maintaining strong performance in automatic speech recognition tasks.

Contribution

The paper provides theoretical and empirical stability conditions for Li-GRU and proposes SLi-GRU, a new variant with five times faster training and improved robustness.

Findings

SLi-GRU trains five times faster than Li-GRU.

SLi-GRU achieves lower word error rates in ASR datasets.

Theoretical stability conditions improve training robustness.

Abstract

The light gated recurrent units (Li-GRU) is well-known for achieving impressive results in automatic speech recognition (ASR) tasks while being lighter and faster to train than a standard gated recurrent units (GRU). However, the unbounded nature of its rectified linear unit on the candidate recurrent gate induces an important gradient exploding phenomenon disrupting the training process and preventing it from being applied to famous datasets. In this paper, we theoretically and empirically derive the necessary conditions for its stability as well as engineering mechanisms to speed up by a factor of five its training time, hence introducing a novel version of this architecture named SLi-GRU. Then, we evaluate its performance both on a toy task illustrating its newly acquired capabilities and a set of three different ASR datasets demonstrating lower word error rates compared to more complex recurrent neural networks.