Decoupling recognition and transcription in Mandarin ASR

TL;DR

This paper proposes a two-step approach for Mandarin ASR by decoupling recognition and transcription, achieving state-of-the-art accuracy on the Aishell-1 dataset.

Contribution

It introduces a novel factorization of Mandarin ASR into audio-to-Pinyin and Pinyin-to-Hanzi tasks, improving recognition accuracy.

Findings

Achieved 3.9% CER on Aishell-1 dataset

Outperforms previous end-to-end methods

Demonstrates effectiveness of decoupling recognition and transcription

Abstract

Much of the recent literature on automatic speech recognition (ASR) is taking an end-to-end approach. Unlike English where the writing system is closely related to sound, Chinese characters (Hanzi) represent meaning, not sound. We propose factoring audio -> Hanzi into two sub-tasks: (1) audio -> Pinyin and (2) Pinyin -> Hanzi, where Pinyin is a system of phonetic transcription of standard Chinese. Factoring the audio -> Hanzi task in this way achieves 3.9% CER (character error rate) on the Aishell-1 corpus, the best result reported on this dataset so far.