Transformer in action: a comparative study of transformer-based acoustic models for large scale speech recognition applications
This work addresses speech recognition efficiency and accuracy for industrial applications, presenting incremental improvements over existing methods.
The paper compares transformer-based acoustic models, including Emformer, against LSTM variants for large-scale speech recognition, showing Emformer achieves 24-26% relative word error rate reductions in low-latency tasks and significant improvements with 2-3 times faster inference in medium-latency scenarios.
In this paper, we summarize the application of transformer and its streamable variant, Emformer based acoustic model for large scale speech recognition applications. We compare the transformer based acoustic models with their LSTM counterparts on industrial scale tasks. Specifically, we compare Emformer with latency-controlled BLSTM (LCBLSTM) on medium latency tasks and LSTM on low latency tasks. On a low latency voice assistant task, Emformer gets 24% to 26% relative word error rate reductions (WERRs). For medium latency scenarios, comparing with LCBLSTM with similar model size and latency, Emformer gets significant WERR across four languages in video captioning datasets with 2-3 times inference real-time factors reduction.