We propose gated language experts and curriculum training to enhance multilingual transformer transducer models without requiring language identification (LID) input from users during inference. Our method incorporates a gating mechanism and LID loss, enabling transformer experts to learn language-specific information. By combining gated transformer experts with shared transformer layers, we construct multilingual transformer blocks and utilize linear experts to effectively regularize the joint network. The curriculum training scheme leverages LID to guide the gated experts in improving their respective language performance. Experimental results on a bilingual task involving English and Spanish demonstrate significant improvements, with average relative word error reductions of 12.5% and 7.3% compared to the baseline bilingual and monolingual models, respectively. Notably, our method achieves performance comparable to the upper-bound model trained and inferred with oracle LID. Extending our approach to trilingual, quadrilingual, and pentalingual models reveals similar advantages to those observed in the bilingual models, highlighting its ease of extension to multiple languages.
@article{arxiv.2303.00786,
title = {Building High-accuracy Multilingual ASR with Gated Language Experts and Curriculum Training},
author = {Eric Sun and Jinyu Li and Yuxuan Hu and Yimeng Zhu and Long Zhou and Jian Xue and Peidong Wang and Linquan Liu and Shujie Liu and Edward Lin and Yifan Gong},
journal= {arXiv preprint arXiv:2303.00786},
year = {2023}
}