Statistical Error Correction Methods for Domain-Specific ASR Systems

Statistical Error Correction Methods for Domain-Specific ASR Systems
复制标题

特定领域 ASR 系统的统计纠错方法

DOI:
--
复制
发表时间:
2013
期刊:
International Conference on Statistical Language and Speech Processing
影响因子:
--
通讯作者:
C. Burileanu
C. Burileanu
中科院分区:
--
文献类型:
--
作者:
H. Cucu;Andi Buzo;L. Besacier;C. Burileanu

文献摘要

被引文献

相似文献

每当 ASR 公司承诺向最终用户提供防错转录本时,就无法避免对原始 ASR 转录本进行手动验证和更正。这种手动后期编辑过程会系统地生成新的、正确的特定领域数据,这些数据可用于逐步改进原始 ASR 系统。本文提出了一种基于 SMT 的统计 ASR 纠错方法,该方法利用过去纠正的 ASR 错误来自动对其未来的成绩单进行后处理。我们表明,仅使用 2000 个用户更正的句子,所提出的方法就可以带来超过 10% 的 WER 改进。
Whenever an ASR company promises to deliver error-proof transcripts to the end user, manual verification and correction of the raw ASR transcripts cannot be avoided. This manual post-editing process systematically generates new and correct domain-specific data which can be used to incrementally improve the original ASR system. This paper proposes a statistic, SMT-based ASR error correction method, which takes advantage of the past corrected ASR errors to automatically post-process its future transcripts. We show that the proposed method can bring more than 10% WER improvements using only 2000 user-corrected sentences.