Human Language Resources for Research in Multilanguage Systems, Robust Recognition, and Speaker Identification

拟议的工作创造了人类语言资源，以刺激三个重要领域的研究：(a)自动语言识别——识别正在使用的语言的问题；(b)鲁棒语音识别——识别来自不同麦克风、通信渠道和环境条件的语音的问题；(c)说话人识别——根据说话人的声音特征识别说话人的问题。为了支持自动语言识别的研究，收集了22种语言中300名发言者的两分钟以上的讲话。为了支持强大识别技术的研究，研究人员从数千名说话者那里收集了大约两分钟的蜂窝语音。为了支持说话人识别的研究，1200名说话人在两年的时间里提供了10次电话通话的语音样本。对于每个语料库，语音按正字法注释，并且每个语料库的一部分按语音转录。这项工作满足了对注释语音语料库的迫切需求：刺激人类语言技术三个重要领域的研究，从而使任何使用一种语言的人都能安全地进入信息高速公路。

The proposed work creates human language resources to stimulate research in three important areas: (a) automatic language identification-- the problem of identifying the language being spoken; (b) robust speech recognition-- the problem of recognizing speech from different microphones, communication channels, and environmental conditions; and (c) speaker recognition-- the problem of identifying a speaker based on his or her voice characteristics. To support research in automatic language identification, over two minutes of speech from 300 speakers in each of twenty two language is being collected. To support research in robust recognition, about two minutes of cellular speech from several thousand speakers is being collected. To support research in speaker recognition, twelve hundred speakers are providing speech samples during ten telephone calls over a two year period. For each corpus, the speech is annotated orthographically, and a portion of each corpus is transcribed phonetically. A critical need for annotated speech corpora is met by this work: to stimulate research in three important areas of human language technology, leading to secure access to the information highway by any person who speaks a language.