To Distill or Not to Distill? On the Robustness of Robust Knowledge Distillation

TL;DR

This paper explores the effectiveness of knowledge distillation in improving Arabic speech recognition models, introducing a new dialectal dataset and demonstrating that distilled models can outperform larger, state-of-the-art models on dialectal and standard benchmarks.

Contribution

It presents a novel Arabic speech recognition dataset for dialects and evaluates a distilled model that surpasses larger models in accuracy and efficiency.

Findings

Distilled model achieves 45.0% WER, outperforming larger models.

New dialectal dataset improves evaluation of ASR models on under-represented dialects.

Error analysis reveals common mistakes and challenges in dialectal speech recognition.

Abstract

Arabic is known to present unique challenges for Automatic Speech Recognition (ASR). On one hand, its rich linguistic diversity and wide range of dialects complicate the development of robust, inclusive models. On the other, current multilingual ASR models are compute-intensive and lack proper comprehensive evaluations. In light of these challenges, we distill knowledge from large teacher models into smaller student variants that are more efficient. We also introduce a novel human-annotated dataset covering five under-represented Arabic dialects for evaluation. We further evaluate both our models and existing SoTA multilingual models on both standard available benchmarks and our new dialectal data. Our best-distilled model's overall performance ($45.0$\% WER) surpasses that of a SoTA model twice its size (SeamlessM4T-large-v2, WER=$47.0$\%) and its teacher model (Whisper-large-v2, WER=$55.1$\%), and its average performance on our new dialectal data ($56.9$\% WER) outperforms all other models. To gain more insight into the poor performance of these models on dialectal data, we conduct an error analysis and report the main types of errors the different models tend to make. The GitHub repository for the project is available at \url{https://github.com/UBC-NLP/distill-whisper-ar}.