Building a great multi-lingual teacher with sparsely-gated mixture of experts for speech recognition
This work addresses improving speech recognition accuracy for multi-lingual applications, representing an incremental advancement by applying an existing technique to specific network architectures.
The paper tackled scaling multi-lingual automatic speech recognition networks using sparsely-gated mixture of experts, achieving relative word error rate reductions of 16.3% for Sequence-to-Sequence Transformer and 4.6% for Transformer Transducer.
The sparsely-gated Mixture of Experts (MoE) can magnify a network capacity with a little computational complexity. In this work, we investigate how multi-lingual Automatic Speech Recognition (ASR) networks can be scaled up with a simple routing algorithm in order to achieve better accuracy. More specifically, we apply the sparsely-gated MoE technique to two types of networks: Sequence-to-Sequence Transformer (S2S-T) and Transformer Transducer (T-T). We demonstrate through a set of ASR experiments on multiple language data that the MoE networks can reduce the relative word error rates by 16.3% and 4.6% with the S2S-T and T-T, respectively. Moreover, we thoroughly investigate the effect of the MoE on the T-T architecture in various conditions: streaming mode, non-streaming mode, the use of language ID and the label decoder with the MoE.