"CU-move" : analysis & corpus development for interactive in-vehicle speech systems
"CU-move" : analysis & corpus development for interactive in-vehicle speech systems
复制标题
“CU-move”：分析
DOI：
10.21437/eurospeech.2001-477
复制
发表时间：
2001
期刊：
影响因子：
--
通讯作者：
R. Cole
中科院分区：
文献类型：
--
作者：
J. Hansen;Pongtep Angkititrakul;Jay P. Plucienkowski;S. Gallant;Umit Yapanel;B. Pellom;Wayne H. Ward;R. Cole
在本文中，我们提出了我们最近的工作，在分析和制定一个新的声学语音语料库开发车载交互系统的路线规划和导航。CU-Move语料库的开发分为两个阶段：[I]跨车辆的声学噪声收集和分析，以及[II]由来自美国各地的1000多名扬声器组成的数据收集。我们目前的结果，从第一阶段的车辆噪声数据分析，以确定第二阶段的大规模数据收集使用一个单一的车辆类型的指导方针。共确定了14种噪声条件，用于分析6种车辆。我们还提出了我们的第二阶段收集计划，包括扬声器，方言区域，数据收集硬件，提示和对话域。由于以前的研究表明，语音识别的性能显着损失时，扬声器是在任务压力下，重要的是要开发会话系统，最大限度地减少操作员的压力，为司机。这将是第一个基于美国的语料库，它的种类包括多通道数据，旨在用于开发混合主动对话语音系统;最初的应用程序是通过连接到WWW的无线信息检索子系统的路线规划和导航。
In this paper, we present our recent work in the analysis and formulation of a new acoustic speech corpus for developing in-vehicle interactive systems for route planning and navigation. The CU-Move Corpus development is partitioned into two phases: [I] acoustic noise collection and analysis across vehicles, and [II] data collection consisting of +1000 speakers from across the United States. We present results from Phase I acoustic noise data analysis across vehicles to determine guidelines for Phase II large-scale data collection using a single vehicle type. A total of 14 noise conditions are identified for analysis across 6 vehicles. We also present our plan for Phase II collection including speakers, dialect regions, data collection hardware, prompts and dialog domains. Since previous studies in speech recognition have shown significant loses in performance when speakers are under task stress, it is important to develop conversational systems that minimize operator stress for the driver. This will be the first U.S. based corpus of its kind consisting of multichannel data, intended for use in developing mixed-initiative dialog speech systems; the initial application being route planning and navigation through a wireless information retrieval sub-system connected to the WWW.