Leave No Knowledge Behind During Knowledge Distillation: Towards Practical and Effective Knowledge Distillation for Code-Switching ASR Using Realistic Data

TL;DR

This paper introduces K²D, a knowledge distillation framework that creates smaller, faster, and more effective code-switching ASR models using realistic data and auxiliary insights, outperforming baselines.

Contribution

The paper proposes K²D, a novel knowledge distillation method that leverages both teacher knowledge and auxiliary models for practical code-switching ASR.

Findings

K²D produces models twice as small and five times faster.

K²D outperforms baseline methods and the teacher model.

The approach is validated on multiple in-domain and out-domain datasets.

Abstract

Recent advances in automatic speech recognition (ASR) often rely on large speech foundation models for generating high-quality transcriptions. However, these models can be impractical due to limited computing resources. The situation is even more severe in terms of more realistic or difficult scenarios, such as code-switching ASR (CS-ASR). To address this, we present a framework for developing more efficient models for CS-ASR through knowledge distillation using realistic speech-only data. Our proposed method, Leave No Knowledge Behind During Knowledge Distillation (K$^2$D), leverages both the teacher model's knowledge and additional insights from a small auxiliary model. We evaluate our approach on two in-domain and two out-domain datasets, demonstrating that K$^2$D is effective. By conducting K$^2$D on the unlabeled realistic data, we have successfully obtained a 2-time smaller model with 5-time faster generation speed while outperforming the baseline methods and the teacher model on all the testing sets. We have made our model publicly available on Hugging Face (https://huggingface.co/andybi7676/k2d-whisper.zh-en).