Knowledge Transfer from Pre-trained Language Models to Cif-based Speech Recognizers via Hierarchical Distillation

TL;DR

This paper introduces hierarchical knowledge distillation to transfer knowledge from pre-trained language models to CIF-based speech recognizers, significantly improving error rates on standard datasets.

Contribution

It proposes a novel hierarchical distillation framework combining acoustic and linguistic knowledge transfer for CIF-based ASR models.

Findings

Achieves 15% relative error reduction on AISHELL-1

Achieves 9% relative error reduction on LibriSpeech

Demonstrates effective knowledge transfer from PLMs to speech recognizers

Abstract

Large-scale pre-trained language models (PLMs) have shown great potential in natural language processing tasks. Leveraging the capabilities of PLMs to enhance automatic speech recognition (ASR) systems has also emerged as a promising research direction. However, previous works may be limited by the inflexible structures of PLMs and the insufficient utilization of PLMs. To alleviate these problems, we propose the hierarchical knowledge distillation (HKD) on the continuous integrate-and-fire (CIF) based ASR models. To transfer knowledge from PLMs to the ASR models, HKD employs cross-modal knowledge distillation with contrastive loss at the acoustic level and knowledge distillation with regression loss at the linguistic level. Compared with the original CIF-based model, our method achieves 15% and 9% relative error rate reduction on the AISHELL-1 and LibriSpeech datasets, respectively.