BART based semantic correction for Mandarin automatic speech recognition system

TL;DR

This paper introduces a BART-based semantic correction method for Mandarin ASR that significantly reduces character errors and improves recognition quality beyond traditional metrics.

Contribution

It proposes a novel Transformer-based semantic correction approach initialized with pretrained BART, enhancing Mandarin speech recognition accuracy.

Findings

CER reduced by 21.7% relative to baseline

Expert evaluation confirms perceptible quality improvement

Effective on large-scale Mandarin speech dataset

Abstract

Although automatic speech recognition (ASR) systems achieved significantly improvements in recent years, spoken language recognition error occurs which can be easily spotted by human beings. Various language modeling techniques have been developed on post recognition tasks like semantic correction. In this paper, we propose a Transformer based semantic correction method with pretrained BART initialization, Experiments on 10000 hours Mandarin speech dataset show that character error rate (CER) can be effectively reduced by 21.7% relatively compared to our baseline ASR system. Expert evaluation demonstrates that actual improvement of our model surpasses what CER indicates.