Mixture-of-Experts with Intermediate CTC Supervision for Accented Speech Recognition

TL;DR

This paper introduces Moe-Ctc, a Mixture-of-Experts model with intermediate CTC supervision, improving accented speech recognition by promoting expert specialization and generalization, leading to significant WER reductions.

Contribution

The paper proposes Moe-Ctc, a novel Mixture-of-Experts architecture with accent-aware routing and intermediate CTC supervision for robust accented speech recognition.

Findings

Achieves up to 29.3% relative WER reduction on Mcv-Accent benchmark.

Demonstrates improved performance on both seen and unseen accents.

Outperforms strong FastConformer baselines.

Abstract

Accented speech remains a persistent challenge for automatic speech recognition (ASR), as most models are trained on data dominated by a few high-resource English varieties, leading to substantial performance degradation for other accents. Accent-agnostic approaches improve robustness yet struggle with heavily accented or unseen varieties, while accent-specific methods rely on limited and often noisy labels. We introduce Moe-Ctc, a Mixture-of-Experts architecture with intermediate CTC supervision that jointly promotes expert specialization and generalization. During training, accent-aware routing encourages experts to capture accent-specific patterns, which gradually transitions to label-free routing for inference. Each expert is equipped with its own CTC head to align routing with transcription quality, and a routing-augmented loss further stabilizes optimization. Experiments on the Mcv-Accent benchmark demonstrate consistent gains across both seen and unseen accents in low- and high-resource conditions, achieving up to 29.3% relative WER reduction over strong FastConformer baselines.