Web-based language modelling for automatic lecture transcription
Web-based language modelling for automatic lecture transcription
复制标题
基于网络的自动讲座转录语言模型
DOI：
10.21437/interspeech.2007-266
复制
发表时间：
2007
期刊：
影响因子：
--
通讯作者：
R. Baecker
中科院分区：
文献类型：
--
作者：
Cosmin Munteanu;Gerald Penn;R. Baecker
长期以来，大学一直依赖书面文本来分享知识。随着越来越多的讲座在网上提供，这些讲座必须附有文字记录，以便提供与教科书一样的信息获取途径。虽然自动语音识别(ASR)是一种具有成本效益的方法，但它在演讲中的准确性还不能令人满意。改进演讲ASR的一种方法是构建更小的、依赖于主题的语言模型(LMS)，并将它们(通过LM内插或假设空间组合)与通用、大词汇量的LMS相结合。在本文中，我们提出了一种简单的解决方案，与基于组合的方法相比，该方法具有相似或更好的错误率(以及特定主题的fic关键词识别fi阳离子准确率)。我们的方法通过利用讲座幻灯片来收集适合于对会话和特定主题的fic风格的讲座进行建模的网络语料库，从而消除了对两种类型的学习模型的需要。
Universities have long relied on written text to share knowledge. As more lectures are made available on-line, these must be accompanied by textual transcripts in order to provide the same access to information as textbooks. While Automatic Speech Recognition (ASR) is a cost-effective method to deliver transcriptions, its accuracy for lectures is not yet satisfactory. One approach for improving lecture ASR is to build smaller, topic-dependent Language Models (LMs) and combine them (through LM interpolation or hypothesis space combination) with general-purpose, large-vocabulary LMs. In this paper, we propose a simple solution for lecture ASR with similar or better Word Error Rate reductions (as well as topic-specific keyword identification accuracies) than combination-based approaches. Our method eliminates the need for two types of LMs by exploiting the lecture slides to collect a web corpus appropriate for modelling both the conversational and the topic-specific styles of lectures.