BriefGPT.xyz
Jun, 2019
探索音素级别的端到端语音翻译表示
Exploring Phoneme-Level Speech Representations for End-to-End Speech Translation
HTML
PDF
Elizabeth Salesky, Matthias Sperber, Alan W Black
TL;DR
研究采用类音素语音表示替代传统帧级语音特征作为源输入，在端到端的语音翻译任务中，相比传统方法，模型性能有显著提高，同时训练时间减少了60%。
Abstract
Previous work on
end-to-end translation
from speech has primarily used
frame-level features
as speech representations, which creates longer, sparser sequences than text. We show that a naive method to create comp
→