Improving Generalization of Deep Neural Network Acoustic Models with Length Perturbation and N-best Based Label Smoothing

被引：0

作者：

Cui, Xiaodong ^{[1
]}

Saon, George ^{[1
]}

Nagano, Tohru ^{[2
]}

Suzuki, Masayuki ^{[2
]}

Fukuda, Takashi ^{[2
]}

Kingsbury, Brian ^{[1
]}

Kurata, Gakuto ^{[2
]}

机构：

[1] IBM Research AI, IBM T. J. Watson Research Center, United States

[2] IBM Research Tokyo, Japan

来源：

Proceedings of the Annual Conference of the International Speech Communication Association, INTERSPEECH | 2022年 / 2022-September卷

关键词：

D O I：

暂无

中图分类号：

学科分类号：

摘要：

Perturbation techniques - Recurrent neural networks - Speech communication - Speech recognition

引用

页码：2638 / 2642