Keep Decoding Parallel with Effective Knowledge Distillation from Language Models to End-to-end Speech Recognisers
This work addresses the challenge of improving speech recognition accuracy for users by enabling efficient knowledge distillation from language models, though it is incremental as it builds on existing distillation and decoding methods.
The study tackled the problem of effectively integrating language model knowledge into end-to-end speech recognition models without sacrificing parallel decoding speed, achieving better recognition accuracy than shallow fusion of an external language model on the LibriSpeech dataset.
This study presents a novel approach for knowledge distillation (KD) from a BERT teacher model to an automatic speech recognition (ASR) model using intermediate layers. To distil the teacher's knowledge, we use an attention decoder that learns from BERT's token probabilities. Our method shows that language model (LM) information can be more effectively distilled into an ASR model using both the intermediate layers and the final layer. By using the intermediate layers as distillation target, we can more effectively distil LM knowledge into the lower network layers. Using our method, we achieve better recognition accuracy than with shallow fusion of an external LM, allowing us to maintain fast parallel decoding. Experiments on the LibriSpeech dataset demonstrate the effectiveness of our approach in enhancing greedy decoding with connectionist temporal classification (CTC).