BLR-MoE: Boosted Language-Routing Mixture of Experts for Domain-Robust Multilingual E2E ASR

TL;DR

This paper introduces BLR-MoE, an advanced architecture for multilingual end-to-end ASR that reduces language confusion through attention-MoE and improved routing, enhancing robustness in domain-mismatched scenarios.

Contribution

It proposes a novel BLR-MoE architecture with attention-MoE and improved routing techniques to better handle language confusion in multilingual ASR.

Findings

BLR-MoE outperforms previous models on a 10,000-hour dataset.

Attention-MoE reduces language confusion in self-attention.

Expert pruning and router augmentation improve routing robustness.

Abstract

Recently, the Mixture of Expert (MoE) architecture, such as LR-MoE, is often used to alleviate the impact of language confusion on the multilingual ASR (MASR) task. However, it still faces language confusion issues, especially in mismatched domain scenarios. In this paper, we decouple language confusion in LR-MoE into confusion in self-attention and router. To alleviate the language confusion in self-attention, based on LR-MoE, we propose to apply attention-MoE architecture for MASR. In our new architecture, MoE is utilized not only on feed-forward network (FFN) but also on self-attention. In addition, to improve the robustness of the LID-based router on language confusion, we propose expert pruning and router augmentation methods. Combining the above, we get the boosted language-routing MoE (BLR-MoE) architecture. We verify the effectiveness of the proposed BLR-MoE in a 10,000-hour MASR dataset.