arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2108.01129cs.CLcs.SDeess.AS
收藏

普通话ASR的去耦合识别与转录

Decoupling recognition and transcription in Mandarin ASR

Baidu Research USA(百度美国研究院);Johns Hopkins University(约翰斯·霍普金斯大学)

Jiahong Yuan, Xingyu Cai, Dongji Gao, Renjie Zheng, Liang Huang, Kenneth Church

摘要:

最近关于自动语音识别(ASR)的许多文献都采用端到端的方法。不像英语的书写系统与声音密切相关,汉字(汉字)代表的是意义,而不是声音。我们建议将音频->汉字分解为两个子任务:(1)音频->拼音和(2)拼音->汉字,其中拼音是标准汉语的音标系统。以这种方式分解音频->汉字任务在Aishell-1语料库上实现了3.9%的CER(字符出错率),这是迄今为止在该数据集上报告的最好结果。

英文摘要:

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.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑