Large Language Model Should Understand Pinyin for Chinese ASR Error Correction

TL;DR

This paper introduces Pinyin-enhanced generative error correction (PY-GEC) for Chinese ASR, leveraging Pinyin phonetic info and multitask training to improve correction accuracy over text-only methods.

Contribution

It proposes a novel Pinyin-based approach with multitask training for Chinese ASR error correction, demonstrating improved performance and interpretability.

Findings

Outperforms text-only GEC on Aishell-1 and Common Voice datasets.

Increases attention on Pinyin features during correction.

Aligns feature spaces between Pinyin and text representations.

Abstract

Large language models can enhance automatic speech recognition systems through generative error correction. In this paper, we propose Pinyin-enhanced GEC, which leverages Pinyi, the phonetic representation of Mandarin Chinese, as supplementary information to improve Chinese ASR error correction. Our approach only utilizes synthetic errors for training and employs the one-best hypothesis during inference. Additionally, we introduce a multitask training approach involving conversion tasks between Pinyin and text to align their feature spaces. Experiments on the Aishell-1 and the Common Voice datasets demonstrate that our approach consistently outperforms GEC with text-only input. More importantly, we provide intuitive explanations for the effectiveness of PY-GEC and multitask training from two aspects: 1) increased attention weight on Pinyin features; and 2) aligned feature space between Pinyin and text hidden states.