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

Wonjun Lee, Hyounghun Kim, Gary Geunbae Lee · 2026

Read Mixture-of-Experts with Intermediate CTC Supervision for Accented Speech Recognition by Wonjun Lee, Hyounghun Kim, Gary Geunbae Lee (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-01967.txt.

Description

Wonjun Lee and colleagues tackle accented speech recognition, where few variety training degrades. Accent agnostic methods help robustness but struggle with accented speech, while accent specific methods need labels. Moe-Ctc, a Mixture of Experts design with intermediate CTC supervision, promotes specialization and generalization. Accent aware routing captures accent patterns, then goes label free. Each expert carries a CTC head aligning routing with quality, plus a routing loss. Mcv-Accent tests show gains across seen and unseen accents, with up to about twenty nine percent lower WER than FastConformer baselines.

Catalog record

CREATORWonjun Lee, Hyounghun Kim, Gary Geunbae Lee
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE38104
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2602.01967

Browse the full library · About the Tower of Babel