arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2005.13681cs.CLcs.SDeess.AS
收藏

手机功能改善语音翻译

Phone Features Improve Speech Translation

Johns Hopkins University(约翰斯·霍普金斯大学);Carnegie Mellon University(卡内基梅隆大学)

Elizabeth Salesky, Alan W Black

摘要:

语音翻译(ST)的端到端模型比传统的单独的ASR和MT模型级联更紧密地耦合语音识别(ASR)和机器翻译(MT),具有更简单的模型架构和减少错误传播的潜力。它们的性能通常被认为是优越的,尽管在许多情况下情况并非如此。我们在高、中、低资源条件下比较级联和端到端模型,并表明级联仍然是更强的基线。此外,我们介绍了两种方法,将手机功能到ST模型。我们表明,这些功能改善了两种架构,缩小了端到端模型和级联之间的差距,并优于以前的学术工作-在我们的低资源设置上高达9 BLEU。

英文摘要:

End-to-end models for speech translation (ST) more tightly couple speech recognition (ASR) and machine translation (MT) than a traditional cascade of separate ASR and MT models, with simpler model architectures and the potential for reduced error propagation. Their performance is often assumed to be superior, though in many conditions this is not yet the case. We compare cascaded and end-to-end models across high, medium, and low-resource conditions, and show that cascades remain stronger baselines. Further, we introduce two methods to incorporate phone features into ST models. We show that these features improve both architectures, closing the gap between end-to-end models and cascades, and outperforming previous academic work -- by up to 9 BLEU on our low-resource setting.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑