Inter-KD: Intermediate Knowledge Distillation for CTC-Based Automatic Speech Recognition

TL;DR

This paper introduces Inter-KD, a novel knowledge distillation method for CTC-based speech recognition that transfers knowledge to intermediate layers, significantly improving performance without additional language models or data augmentation.

Contribution

The paper proposes Inter-KD, a new intermediate-layer knowledge distillation technique for CTC models, enhancing speech recognition accuracy efficiently.

Findings

Inter-KD outperforms conventional KD methods on LibriSpeech.

Inter-KD reduces WER from 8.85% to 6.30% on test-clean.

No language model or data augmentation needed.

Abstract

Recently, the advance in deep learning has brought a considerable improvement in the end-to-end speech recognition field, simplifying the traditional pipeline while producing promising results. Among the end-to-end models, the connectionist temporal classification (CTC)-based model has attracted research interest due to its non-autoregressive nature. However, such CTC models require a heavy computational cost to achieve outstanding performance. To mitigate the computational burden, we propose a simple yet effective knowledge distillation (KD) for the CTC framework, namely Inter-KD, that additionally transfers the teacher's knowledge to the intermediate CTC layers of the student network. From the experimental results on the LibriSpeech, we verify that the Inter-KD shows better achievements compared to the conventional KD methods. Without using any language model (LM) and data augmentation, Inter-KD improves the word error rate (WER) performance from 8.85 % to 6.30 % on the test-clean.