BriefGPT.xyz
Nov, 2022
探究用于端到端语音识别的统计表示
Probing Statistical Representations For End-To-End ASR
HTML
PDF
Anna Ollerenshaw, Md Asif Jalal, Thomas Hain
TL;DR
分析了transformer架构中跨域语言模型依赖关系的研究,使用SVCCA发现转换器层中的特定神经表示具有相关行为,并影响识别性能。这项工作提供了有关模型方法的分析,这些模型方法影响了环境依赖关系和ASR性能,可以用于创建或调整性能更好的End-to-End ASR模型和下游任务。
Abstract
End-to-End
automatic speech recognition
(ASR) models aim to learn a generalised speech representation to perform recognition. In this domain there is little research to analyse internal representation dependencies and their relationship to
→