arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 1603.09170cs.CLcs.LGstat.ML
收藏

基于跨维随机场语言模型的语音识别模型插值

Model Interpolation with Trans-dimensional Random Field Language Models for Speech Recognition

Tsinghua University(清华大学);Toshiba Corporation(东芝公司)

Bin Wang, Zhijian Ou, Yong He, Akinori Kawamura

摘要:

主要的语言模型(LM),如n-gram和神经网络(NN)模型表示句子概率的条件。相比之下,一个新的跨维随机场(TRF)LM最近已被引入,以显示优越的性能,其中整个句子被建模为一个随机场。在本文中,我们研究了如何将TRF模型与NN模型进行插值,并通过对数线性组合,在6-gram线性模型上分别获得了12.1%和17.9%的相对出错率降低,用于英语和汉语语音识别。

英文摘要:

The dominant language models (LMs) such as n-gram and neural network (NN) models represent sentence probabilities in terms of conditionals. In contrast, a new trans-dimensional random field (TRF) LM has been recently introduced to show superior performances, where the whole sentence is modeled as a random field. In this paper, we examine how the TRF models can be interpolated with the NN models, and obtain 12.1\% and 17.9\% relative error rate reductions over 6-gram LMs for English and Chinese speech recognition respectively through log-linear combination.

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。

↑