Using fine-tuning and min lookahead beam search to improve Whisper

TL;DR

This paper enhances Whisper's performance on low-resource languages by fine-tuning with LoRA and introducing Min Lookahead beam search, significantly reducing WER across multiple languages.

Contribution

It proposes an improved decoding algorithm and demonstrates that fine-tuning with LoRA and Min Lookahead significantly improves Whisper's accuracy.

Findings

38.49 WER reduction on Vietnamese with LoRA fine-tuning

2.26 WER reduction using Min Lookahead over standard beam search

Theoretical proof that Min Lookahead outperforms standard beam search

Abstract

The performance of Whisper in low-resource languages is still far from perfect. In addition to a lack of training data on low-resource languages, we identify some limitations in the beam search algorithm used in Whisper. To address these issues, we fine-tune Whisper on additional data and propose an improved decoding algorithm. On the Vietnamese language, fine-tuning Whisper-Tiny with LoRA leads to an improvement of 38.49 in WER over the zero-shot Whisper-Tiny setting which is a further reduction of 1.45 compared to full-parameter fine-tuning. Additionally, by using Filter-Ends and Min Lookahead decoding algorithms, the WER reduces by 2.26 on average over a range of languages compared to standard beam search. These results generalise to larger Whisper model sizes. We also prove a theorem that Min Lookahead outperforms the standard beam search algorithm used in Whisper.