• 主题
高级搜索  >
历史搜索 清空历史
首页> 外文会议>International Speech Communication Association >Stream Decoding for Simultaneous Spoken Language Translation
【24h】

Stream Decoding for Simultaneous Spoken Language Translation

机译：用于同时口语语言翻译的流解码

获取原文
页面导航
  • 摘要
  • 著录项
  • 相似文献
  • 相关主题

摘要

In the typical speech translation system, the first-bestspeech recognizer hypothesis is segmented into sentence-likeunits which are then fed to the downstream machine transla-tion component. The need for a sufficiently large context inthis intermediate step and for the MT introduces delays whichare undesirable in many application scenarios, such as real-timesubtitling of foreign language broadcasts or simultaneous trans-lation of speeches and lectures. In this paper, we propose a statistical machine translationdecoder which processes a continuous input stream, such as thatproduced by a run-on speech recognizer. By decoupling deci-sions about the timing of translation output generation from anyfixed input segmentation, this design can guarantee a maximumoutput lag for each input word while allowing for full word re-ordering within this time window. Experimental results show that this system achieves com-petitive translation performance with a minimum of translation-induced latency.
机译：在典型的语音翻译系统中，将第一estspeech识别器假设分段为句子状单位，然后将其馈送到下游机器换档组件。需要足够大的上下文Inthis中间步骤和MT的延迟介绍许多应用场景中不希望的延迟，例如外语广播的实际时间，或者同时对演讲和讲座的同时跨越横向的跨越。在本文中，我们提出了一种统计机器翻译，该统计机器翻译器处理连续输入流，例如由运行上的语音识别器产生的。通过从任何固定的输入分割的翻译输出生成的定时解耦的分数，可以保证每个输入字的最终输出滞后，同时在此时间窗口内允许全文重新排序。实验结果表明，该系统达到了相应的翻译性能，最小的翻译引起的延迟。

著录项

相似文献

  • 外文文献
  • 中文文献
  • 专利
获取原文

客服邮箱：kefu@zhangqiaokeyan.com

京公网安备：11010802029741号 ICP备案号：京ICP备15016152号-6 六维联合信息科技 (北京) 有限公司©版权所有

  • 客服微信

  • 服务号