BriefGPT.xyz
Feb, 2020
无监督预训练在语言间具备良好的迁移能力
Unsupervised pretraining transfers well across languages
HTML
PDF
Morgane Rivière, Armand Joulin, Pierre-Emmanuel Mazaré, Emmanuel Dupoux
TL;DR
本篇研究调查了无监督预训练是否能够跨语言传输，以便自动语音识别系统（ASR）实现跨语言和多语言。研究表明，使用略加修改的对比性预测编码（CPC）预训练方式，能够提取和其他语言效果相当或甚至优于监督预训练的特征，证明了无监督方法在语言资源稀缺的情况下具有潜力。
Abstract
Cross-lingual and
multi-lingual training
of
automatic speech recognition
(ASR) has been extensively investigated in the supervised setting. This assumes the existence of a parallel corpus of speech and orthograph
→